1 /*
2  *  Copyright (c) 2012 The WebRTC project authors. All Rights Reserved.
3  *
4  *  Use of this source code is governed by a BSD-style license
5  *  that can be found in the LICENSE file in the root of the source
6  *  tree. An additional intellectual property rights grant can be found
7  *  in the file PATENTS.  All contributing project authors may
8  *  be found in the AUTHORS file in the root of the source tree.
9  */
10 
11 /*
12  * This header file includes all of the fix point signal processing library
13  * (SPL) function descriptions and declarations. For specific function calls,
14  * see bottom of file.
15  */
16 
17 #ifndef COMMON_AUDIO_SIGNAL_PROCESSING_INCLUDE_SIGNAL_PROCESSING_LIBRARY_H_
18 #define COMMON_AUDIO_SIGNAL_PROCESSING_INCLUDE_SIGNAL_PROCESSING_LIBRARY_H_
19 
20 #include <string.h>
21 
22 #include "common_audio/signal_processing/dot_product_with_scale.h"
23 
24 // Macros specific for the fixed point implementation
25 #define WEBRTC_SPL_WORD16_MAX 32767
26 #define WEBRTC_SPL_WORD16_MIN -32768
27 #define WEBRTC_SPL_WORD32_MAX (int32_t)0x7fffffff
28 #define WEBRTC_SPL_WORD32_MIN (int32_t)0x80000000
29 #define WEBRTC_SPL_MAX_LPC_ORDER 14
30 #define WEBRTC_SPL_MIN(A, B) (A < B ? A : B)  // Get min value
31 #define WEBRTC_SPL_MAX(A, B) (A > B ? A : B)  // Get max value
32 // TODO(kma/bjorn): For the next two macros, investigate how to correct the code
33 // for inputs of a = WEBRTC_SPL_WORD16_MIN or WEBRTC_SPL_WORD32_MIN.
34 #define WEBRTC_SPL_ABS_W16(a) (((int16_t)a >= 0) ? ((int16_t)a) : -((int16_t)a))
35 #define WEBRTC_SPL_ABS_W32(a) (((int32_t)a >= 0) ? ((int32_t)a) : -((int32_t)a))
36 
37 #define WEBRTC_SPL_MUL(a, b) ((int32_t)((int32_t)(a) * (int32_t)(b)))
38 #define WEBRTC_SPL_UMUL(a, b) ((uint32_t)((uint32_t)(a) * (uint32_t)(b)))
39 #define WEBRTC_SPL_UMUL_32_16(a, b) ((uint32_t)((uint32_t)(a) * (uint16_t)(b)))
40 #define WEBRTC_SPL_MUL_16_U16(a, b) ((int32_t)(int16_t)(a) * (uint16_t)(b))
41 
42 // clang-format off
43 // clang-format would choose some identation
44 // leading to presubmit error (cpplint.py)
45 #ifndef WEBRTC_ARCH_ARM_V7
46 // For ARMv7 platforms, these are inline functions in spl_inl_armv7.h
47 #ifndef MIPS32_LE
48 // For MIPS platforms, these are inline functions in spl_inl_mips.h
49 #define WEBRTC_SPL_MUL_16_16(a, b) ((int32_t)(((int16_t)(a)) * ((int16_t)(b))))
50 #define WEBRTC_SPL_MUL_16_32_RSFT16(a, b) \
51         (WEBRTC_SPL_MUL_16_16(a, b >> 16) +     \
52         ((WEBRTC_SPL_MUL_16_16(a, (b & 0xffff) >> 1) + 0x4000) >> 15))
53 #endif
54 #endif
55 
56 #define WEBRTC_SPL_MUL_16_32_RSFT11(a, b)          \
57         (WEBRTC_SPL_MUL_16_16(a, (b) >> 16) * (1 << 5) + \
58         (((WEBRTC_SPL_MUL_16_U16(a, (uint16_t)(b)) >> 1) + 0x0200) >> 10))
59 #define WEBRTC_SPL_MUL_16_32_RSFT14(a, b)          \
60         (WEBRTC_SPL_MUL_16_16(a, (b) >> 16) * (1 << 2) + \
61         (((WEBRTC_SPL_MUL_16_U16(a, (uint16_t)(b)) >> 1) + 0x1000) >> 13))
62 #define WEBRTC_SPL_MUL_16_32_RSFT15(a, b)            \
63         ((WEBRTC_SPL_MUL_16_16(a, (b) >> 16) * (1 << 1)) + \
64         (((WEBRTC_SPL_MUL_16_U16(a, (uint16_t)(b)) >> 1) + 0x2000) >> 14))
65 // clang-format on
66 
67 #define WEBRTC_SPL_MUL_16_16_RSFT(a, b, c) (WEBRTC_SPL_MUL_16_16(a, b) >> (c))
68 
69 #define WEBRTC_SPL_MUL_16_16_RSFT_WITH_ROUND(a, b, c) \
70   ((WEBRTC_SPL_MUL_16_16(a, b) + ((int32_t)(((int32_t)1) << ((c)-1)))) >> (c))
71 
72 // C + the 32 most significant bits of A * B
73 #define WEBRTC_SPL_SCALEDIFF32(A, B, C) \
74   (C + (B >> 16) * A + (((uint32_t)(B & 0x0000FFFF) * A) >> 16))
75 
76 #define WEBRTC_SPL_SAT(a, b, c) (b > a ? a : b < c ? c : b)
77 
78 // Shifting with negative numbers allowed
79 // Positive means left shift
80 #define WEBRTC_SPL_SHIFT_W32(x, c) ((c) >= 0 ? (x) * (1 << (c)) : (x) >> -(c))
81 
82 // Shifting with negative numbers not allowed
83 // We cannot do casting here due to signed/unsigned problem
84 #define WEBRTC_SPL_LSHIFT_W32(x, c) ((x) << (c))
85 
86 #define WEBRTC_SPL_RSHIFT_U32(x, c) ((uint32_t)(x) >> (c))
87 
88 #define WEBRTC_SPL_RAND(a) ((int16_t)((((int16_t)a * 18816) >> 7) & 0x00007fff))
89 
90 #ifdef __cplusplus
91 extern "C" {
92 #endif
93 
94 #define WEBRTC_SPL_MEMCPY_W16(v1, v2, length) \
95   memcpy(v1, v2, (length) * sizeof(int16_t))
96 
97 // inline functions:
98 #include "common_audio/signal_processing/include/spl_inl.h"
99 
100 // third party math functions
101 #include "common_audio/third_party/spl_sqrt_floor/spl_sqrt_floor.h"
102 
103 int16_t WebRtcSpl_GetScalingSquare(int16_t* in_vector,
104                                    size_t in_vector_length,
105                                    size_t times);
106 
107 // Copy and set operations. Implementation in copy_set_operations.c.
108 // Descriptions at bottom of file.
109 void WebRtcSpl_MemSetW16(int16_t* vector,
110                          int16_t set_value,
111                          size_t vector_length);
112 void WebRtcSpl_MemSetW32(int32_t* vector,
113                          int32_t set_value,
114                          size_t vector_length);
115 void WebRtcSpl_MemCpyReversedOrder(int16_t* out_vector,
116                                    int16_t* in_vector,
117                                    size_t vector_length);
118 void WebRtcSpl_CopyFromEndW16(const int16_t* in_vector,
119                               size_t in_vector_length,
120                               size_t samples,
121                               int16_t* out_vector);
122 void WebRtcSpl_ZerosArrayW16(int16_t* vector, size_t vector_length);
123 void WebRtcSpl_ZerosArrayW32(int32_t* vector, size_t vector_length);
124 // End: Copy and set operations.
125 
126 // Minimum and maximum operation functions and their pointers.
127 // Implementation in min_max_operations.c.
128 
129 // Returns the largest absolute value in a signed 16-bit vector.
130 //
131 // Input:
132 //      - vector : 16-bit input vector.
133 //      - length : Number of samples in vector.
134 //
135 // Return value  : Maximum absolute value in vector.
136 typedef int16_t (*MaxAbsValueW16)(const int16_t* vector, size_t length);
137 extern const MaxAbsValueW16 WebRtcSpl_MaxAbsValueW16;
138 int16_t WebRtcSpl_MaxAbsValueW16C(const int16_t* vector, size_t length);
139 #if defined(WEBRTC_HAS_NEON)
140 int16_t WebRtcSpl_MaxAbsValueW16Neon(const int16_t* vector, size_t length);
141 #endif
142 #if defined(MIPS32_LE)
143 int16_t WebRtcSpl_MaxAbsValueW16_mips(const int16_t* vector, size_t length);
144 #endif
145 
146 // Returns the largest absolute value in a signed 32-bit vector.
147 //
148 // Input:
149 //      - vector : 32-bit input vector.
150 //      - length : Number of samples in vector.
151 //
152 // Return value  : Maximum absolute value in vector.
153 typedef int32_t (*MaxAbsValueW32)(const int32_t* vector, size_t length);
154 extern const MaxAbsValueW32 WebRtcSpl_MaxAbsValueW32;
155 int32_t WebRtcSpl_MaxAbsValueW32C(const int32_t* vector, size_t length);
156 #if defined(WEBRTC_HAS_NEON)
157 int32_t WebRtcSpl_MaxAbsValueW32Neon(const int32_t* vector, size_t length);
158 #endif
159 #if defined(MIPS_DSP_R1_LE)
160 int32_t WebRtcSpl_MaxAbsValueW32_mips(const int32_t* vector, size_t length);
161 #endif
162 
163 // Returns the maximum value of a 16-bit vector.
164 //
165 // Input:
166 //      - vector : 16-bit input vector.
167 //      - length : Number of samples in vector.
168 //
169 // Return value  : Maximum sample value in |vector|.
170 typedef int16_t (*MaxValueW16)(const int16_t* vector, size_t length);
171 extern const MaxValueW16 WebRtcSpl_MaxValueW16;
172 int16_t WebRtcSpl_MaxValueW16C(const int16_t* vector, size_t length);
173 #if defined(WEBRTC_HAS_NEON)
174 int16_t WebRtcSpl_MaxValueW16Neon(const int16_t* vector, size_t length);
175 #endif
176 #if defined(MIPS32_LE)
177 int16_t WebRtcSpl_MaxValueW16_mips(const int16_t* vector, size_t length);
178 #endif
179 
180 // Returns the maximum value of a 32-bit vector.
181 //
182 // Input:
183 //      - vector : 32-bit input vector.
184 //      - length : Number of samples in vector.
185 //
186 // Return value  : Maximum sample value in |vector|.
187 typedef int32_t (*MaxValueW32)(const int32_t* vector, size_t length);
188 extern const MaxValueW32 WebRtcSpl_MaxValueW32;
189 int32_t WebRtcSpl_MaxValueW32C(const int32_t* vector, size_t length);
190 #if defined(WEBRTC_HAS_NEON)
191 int32_t WebRtcSpl_MaxValueW32Neon(const int32_t* vector, size_t length);
192 #endif
193 #if defined(MIPS32_LE)
194 int32_t WebRtcSpl_MaxValueW32_mips(const int32_t* vector, size_t length);
195 #endif
196 
197 // Returns the minimum value of a 16-bit vector.
198 //
199 // Input:
200 //      - vector : 16-bit input vector.
201 //      - length : Number of samples in vector.
202 //
203 // Return value  : Minimum sample value in |vector|.
204 typedef int16_t (*MinValueW16)(const int16_t* vector, size_t length);
205 extern const MinValueW16 WebRtcSpl_MinValueW16;
206 int16_t WebRtcSpl_MinValueW16C(const int16_t* vector, size_t length);
207 #if defined(WEBRTC_HAS_NEON)
208 int16_t WebRtcSpl_MinValueW16Neon(const int16_t* vector, size_t length);
209 #endif
210 #if defined(MIPS32_LE)
211 int16_t WebRtcSpl_MinValueW16_mips(const int16_t* vector, size_t length);
212 #endif
213 
214 // Returns the minimum value of a 32-bit vector.
215 //
216 // Input:
217 //      - vector : 32-bit input vector.
218 //      - length : Number of samples in vector.
219 //
220 // Return value  : Minimum sample value in |vector|.
221 typedef int32_t (*MinValueW32)(const int32_t* vector, size_t length);
222 extern const MinValueW32 WebRtcSpl_MinValueW32;
223 int32_t WebRtcSpl_MinValueW32C(const int32_t* vector, size_t length);
224 #if defined(WEBRTC_HAS_NEON)
225 int32_t WebRtcSpl_MinValueW32Neon(const int32_t* vector, size_t length);
226 #endif
227 #if defined(MIPS32_LE)
228 int32_t WebRtcSpl_MinValueW32_mips(const int32_t* vector, size_t length);
229 #endif
230 
231 // Returns the vector index to the largest absolute value of a 16-bit vector.
232 //
233 // Input:
234 //      - vector : 16-bit input vector.
235 //      - length : Number of samples in vector.
236 //
237 // Return value  : Index to the maximum absolute value in vector.
238 //                 If there are multiple equal maxima, return the index of the
239 //                 first. -32768 will always have precedence over 32767 (despite
240 //                 -32768 presenting an int16 absolute value of 32767).
241 size_t WebRtcSpl_MaxAbsIndexW16(const int16_t* vector, size_t length);
242 
243 // Returns the vector index to the maximum sample value of a 16-bit vector.
244 //
245 // Input:
246 //      - vector : 16-bit input vector.
247 //      - length : Number of samples in vector.
248 //
249 // Return value  : Index to the maximum value in vector (if multiple
250 //                 indexes have the maximum, return the first).
251 size_t WebRtcSpl_MaxIndexW16(const int16_t* vector, size_t length);
252 
253 // Returns the vector index to the maximum sample value of a 32-bit vector.
254 //
255 // Input:
256 //      - vector : 32-bit input vector.
257 //      - length : Number of samples in vector.
258 //
259 // Return value  : Index to the maximum value in vector (if multiple
260 //                 indexes have the maximum, return the first).
261 size_t WebRtcSpl_MaxIndexW32(const int32_t* vector, size_t length);
262 
263 // Returns the vector index to the minimum sample value of a 16-bit vector.
264 //
265 // Input:
266 //      - vector : 16-bit input vector.
267 //      - length : Number of samples in vector.
268 //
269 // Return value  : Index to the mimimum value in vector  (if multiple
270 //                 indexes have the minimum, return the first).
271 size_t WebRtcSpl_MinIndexW16(const int16_t* vector, size_t length);
272 
273 // Returns the vector index to the minimum sample value of a 32-bit vector.
274 //
275 // Input:
276 //      - vector : 32-bit input vector.
277 //      - length : Number of samples in vector.
278 //
279 // Return value  : Index to the mimimum value in vector  (if multiple
280 //                 indexes have the minimum, return the first).
281 size_t WebRtcSpl_MinIndexW32(const int32_t* vector, size_t length);
282 
283 // End: Minimum and maximum operations.
284 
285 // Vector scaling operations. Implementation in vector_scaling_operations.c.
286 // Description at bottom of file.
287 void WebRtcSpl_VectorBitShiftW16(int16_t* out_vector,
288                                  size_t vector_length,
289                                  const int16_t* in_vector,
290                                  int16_t right_shifts);
291 void WebRtcSpl_VectorBitShiftW32(int32_t* out_vector,
292                                  size_t vector_length,
293                                  const int32_t* in_vector,
294                                  int16_t right_shifts);
295 void WebRtcSpl_VectorBitShiftW32ToW16(int16_t* out_vector,
296                                       size_t vector_length,
297                                       const int32_t* in_vector,
298                                       int right_shifts);
299 void WebRtcSpl_ScaleVector(const int16_t* in_vector,
300                            int16_t* out_vector,
301                            int16_t gain,
302                            size_t vector_length,
303                            int16_t right_shifts);
304 void WebRtcSpl_ScaleVectorWithSat(const int16_t* in_vector,
305                                   int16_t* out_vector,
306                                   int16_t gain,
307                                   size_t vector_length,
308                                   int16_t right_shifts);
309 void WebRtcSpl_ScaleAndAddVectors(const int16_t* in_vector1,
310                                   int16_t gain1,
311                                   int right_shifts1,
312                                   const int16_t* in_vector2,
313                                   int16_t gain2,
314                                   int right_shifts2,
315                                   int16_t* out_vector,
316                                   size_t vector_length);
317 
318 // The functions (with related pointer) perform the vector operation:
319 //   out_vector[k] = ((scale1 * in_vector1[k]) + (scale2 * in_vector2[k])
320 //        + round_value) >> right_shifts,
321 //   where  round_value = (1 << right_shifts) >> 1.
322 //
323 // Input:
324 //      - in_vector1       : Input vector 1
325 //      - in_vector1_scale : Gain to be used for vector 1
326 //      - in_vector2       : Input vector 2
327 //      - in_vector2_scale : Gain to be used for vector 2
328 //      - right_shifts     : Number of right bit shifts to be applied
329 //      - length           : Number of elements in the input vectors
330 //
331 // Output:
332 //      - out_vector       : Output vector
333 // Return value            : 0 if OK, -1 if (in_vector1 == null
334 //                           || in_vector2 == null || out_vector == null
335 //                           || length <= 0 || right_shift < 0).
336 typedef int (*ScaleAndAddVectorsWithRound)(const int16_t* in_vector1,
337                                            int16_t in_vector1_scale,
338                                            const int16_t* in_vector2,
339                                            int16_t in_vector2_scale,
340                                            int right_shifts,
341                                            int16_t* out_vector,
342                                            size_t length);
343 extern const ScaleAndAddVectorsWithRound WebRtcSpl_ScaleAndAddVectorsWithRound;
344 int WebRtcSpl_ScaleAndAddVectorsWithRoundC(const int16_t* in_vector1,
345                                            int16_t in_vector1_scale,
346                                            const int16_t* in_vector2,
347                                            int16_t in_vector2_scale,
348                                            int right_shifts,
349                                            int16_t* out_vector,
350                                            size_t length);
351 #if defined(MIPS_DSP_R1_LE)
352 int WebRtcSpl_ScaleAndAddVectorsWithRound_mips(const int16_t* in_vector1,
353                                                int16_t in_vector1_scale,
354                                                const int16_t* in_vector2,
355                                                int16_t in_vector2_scale,
356                                                int right_shifts,
357                                                int16_t* out_vector,
358                                                size_t length);
359 #endif
360 // End: Vector scaling operations.
361 
362 // iLBC specific functions. Implementations in ilbc_specific_functions.c.
363 // Description at bottom of file.
364 void WebRtcSpl_ReverseOrderMultArrayElements(int16_t* out_vector,
365                                              const int16_t* in_vector,
366                                              const int16_t* window,
367                                              size_t vector_length,
368                                              int16_t right_shifts);
369 void WebRtcSpl_ElementwiseVectorMult(int16_t* out_vector,
370                                      const int16_t* in_vector,
371                                      const int16_t* window,
372                                      size_t vector_length,
373                                      int16_t right_shifts);
374 void WebRtcSpl_AddVectorsAndShift(int16_t* out_vector,
375                                   const int16_t* in_vector1,
376                                   const int16_t* in_vector2,
377                                   size_t vector_length,
378                                   int16_t right_shifts);
379 void WebRtcSpl_AddAffineVectorToVector(int16_t* out_vector,
380                                        const int16_t* in_vector,
381                                        int16_t gain,
382                                        int32_t add_constant,
383                                        int16_t right_shifts,
384                                        size_t vector_length);
385 void WebRtcSpl_AffineTransformVector(int16_t* out_vector,
386                                      const int16_t* in_vector,
387                                      int16_t gain,
388                                      int32_t add_constant,
389                                      int16_t right_shifts,
390                                      size_t vector_length);
391 // End: iLBC specific functions.
392 
393 // Signal processing operations.
394 
395 // A 32-bit fix-point implementation of auto-correlation computation
396 //
397 // Input:
398 //      - in_vector        : Vector to calculate autocorrelation upon
399 //      - in_vector_length : Length (in samples) of |vector|
400 //      - order            : The order up to which the autocorrelation should be
401 //                           calculated
402 //
403 // Output:
404 //      - result           : auto-correlation values (values should be seen
405 //                           relative to each other since the absolute values
406 //                           might have been down shifted to avoid overflow)
407 //
408 //      - scale            : The number of left shifts required to obtain the
409 //                           auto-correlation in Q0
410 //
411 // Return value            : Number of samples in |result|, i.e. (order+1)
412 size_t WebRtcSpl_AutoCorrelation(const int16_t* in_vector,
413                                  size_t in_vector_length,
414                                  size_t order,
415                                  int32_t* result,
416                                  int* scale);
417 
418 // A 32-bit fix-point implementation of the Levinson-Durbin algorithm that
419 // does NOT use the 64 bit class
420 //
421 // Input:
422 //      - auto_corr : Vector with autocorrelation values of length >= |order|+1
423 //      - order     : The LPC filter order (support up to order 20)
424 //
425 // Output:
426 //      - lpc_coef  : lpc_coef[0..order] LPC coefficients in Q12
427 //      - refl_coef : refl_coef[0...order-1]| Reflection coefficients in Q15
428 //
429 // Return value     : 1 for stable 0 for unstable
430 int16_t WebRtcSpl_LevinsonDurbin(const int32_t* auto_corr,
431                                  int16_t* lpc_coef,
432                                  int16_t* refl_coef,
433                                  size_t order);
434 
435 // Converts reflection coefficients |refl_coef| to LPC coefficients |lpc_coef|.
436 // This version is a 16 bit operation.
437 //
438 // NOTE: The 16 bit refl_coef -> lpc_coef conversion might result in a
439 // "slightly unstable" filter (i.e., a pole just outside the unit circle) in
440 // "rare" cases even if the reflection coefficients are stable.
441 //
442 // Input:
443 //      - refl_coef : Reflection coefficients in Q15 that should be converted
444 //                    to LPC coefficients
445 //      - use_order : Number of coefficients in |refl_coef|
446 //
447 // Output:
448 //      - lpc_coef  : LPC coefficients in Q12
449 void WebRtcSpl_ReflCoefToLpc(const int16_t* refl_coef,
450                              int use_order,
451                              int16_t* lpc_coef);
452 
453 // Converts LPC coefficients |lpc_coef| to reflection coefficients |refl_coef|.
454 // This version is a 16 bit operation.
455 // The conversion is implemented by the step-down algorithm.
456 //
457 // Input:
458 //      - lpc_coef  : LPC coefficients in Q12, that should be converted to
459 //                    reflection coefficients
460 //      - use_order : Number of coefficients in |lpc_coef|
461 //
462 // Output:
463 //      - refl_coef : Reflection coefficients in Q15.
464 void WebRtcSpl_LpcToReflCoef(int16_t* lpc_coef,
465                              int use_order,
466                              int16_t* refl_coef);
467 
468 // Calculates reflection coefficients (16 bit) from auto-correlation values
469 //
470 // Input:
471 //      - auto_corr : Auto-correlation values
472 //      - use_order : Number of coefficients wanted be calculated
473 //
474 // Output:
475 //      - refl_coef : Reflection coefficients in Q15.
476 void WebRtcSpl_AutoCorrToReflCoef(const int32_t* auto_corr,
477                                   int use_order,
478                                   int16_t* refl_coef);
479 
480 // The functions (with related pointer) calculate the cross-correlation between
481 // two sequences |seq1| and |seq2|.
482 // |seq1| is fixed and |seq2| slides as the pointer is increased with the
483 // amount |step_seq2|. Note the arguments should obey the relationship:
484 // |dim_seq| - 1 + |step_seq2| * (|dim_cross_correlation| - 1) <
485 //      buffer size of |seq2|
486 //
487 // Input:
488 //      - seq1           : First sequence (fixed throughout the correlation)
489 //      - seq2           : Second sequence (slides |step_vector2| for each
490 //                            new correlation)
491 //      - dim_seq        : Number of samples to use in the cross-correlation
492 //      - dim_cross_correlation : Number of cross-correlations to calculate (the
493 //                            start position for |vector2| is updated for each
494 //                            new one)
495 //      - right_shifts   : Number of right bit shifts to use. This will
496 //                            become the output Q-domain.
497 //      - step_seq2      : How many (positive or negative) steps the
498 //                            |vector2| pointer should be updated for each new
499 //                            cross-correlation value.
500 //
501 // Output:
502 //      - cross_correlation : The cross-correlation in Q(-right_shifts)
503 typedef void (*CrossCorrelation)(int32_t* cross_correlation,
504                                  const int16_t* seq1,
505                                  const int16_t* seq2,
506                                  size_t dim_seq,
507                                  size_t dim_cross_correlation,
508                                  int right_shifts,
509                                  int step_seq2);
510 extern const CrossCorrelation WebRtcSpl_CrossCorrelation;
511 void WebRtcSpl_CrossCorrelationC(int32_t* cross_correlation,
512                                  const int16_t* seq1,
513                                  const int16_t* seq2,
514                                  size_t dim_seq,
515                                  size_t dim_cross_correlation,
516                                  int right_shifts,
517                                  int step_seq2);
518 #if defined(WEBRTC_HAS_NEON)
519 void WebRtcSpl_CrossCorrelationNeon(int32_t* cross_correlation,
520                                     const int16_t* seq1,
521                                     const int16_t* seq2,
522                                     size_t dim_seq,
523                                     size_t dim_cross_correlation,
524                                     int right_shifts,
525                                     int step_seq2);
526 #endif
527 #if defined(MIPS32_LE)
528 void WebRtcSpl_CrossCorrelation_mips(int32_t* cross_correlation,
529                                      const int16_t* seq1,
530                                      const int16_t* seq2,
531                                      size_t dim_seq,
532                                      size_t dim_cross_correlation,
533                                      int right_shifts,
534                                      int step_seq2);
535 #endif
536 
537 // Creates (the first half of) a Hanning window. Size must be at least 1 and
538 // at most 512.
539 //
540 // Input:
541 //      - size      : Length of the requested Hanning window (1 to 512)
542 //
543 // Output:
544 //      - window    : Hanning vector in Q14.
545 void WebRtcSpl_GetHanningWindow(int16_t* window, size_t size);
546 
547 // Calculates y[k] = sqrt(1 - x[k]^2) for each element of the input vector
548 // |in_vector|. Input and output values are in Q15.
549 //
550 // Inputs:
551 //      - in_vector     : Values to calculate sqrt(1 - x^2) of
552 //      - vector_length : Length of vector |in_vector|
553 //
554 // Output:
555 //      - out_vector    : Output values in Q15
556 void WebRtcSpl_SqrtOfOneMinusXSquared(int16_t* in_vector,
557                                       size_t vector_length,
558                                       int16_t* out_vector);
559 // End: Signal processing operations.
560 
561 // Randomization functions. Implementations collected in
562 // randomization_functions.c and descriptions at bottom of this file.
563 int16_t WebRtcSpl_RandU(uint32_t* seed);
564 int16_t WebRtcSpl_RandN(uint32_t* seed);
565 int16_t WebRtcSpl_RandUArray(int16_t* vector,
566                              int16_t vector_length,
567                              uint32_t* seed);
568 // End: Randomization functions.
569 
570 // Math functions
571 int32_t WebRtcSpl_Sqrt(int32_t value);
572 
573 // Divisions. Implementations collected in division_operations.c and
574 // descriptions at bottom of this file.
575 uint32_t WebRtcSpl_DivU32U16(uint32_t num, uint16_t den);
576 int32_t WebRtcSpl_DivW32W16(int32_t num, int16_t den);
577 int16_t WebRtcSpl_DivW32W16ResW16(int32_t num, int16_t den);
578 int32_t WebRtcSpl_DivResultInQ31(int32_t num, int32_t den);
579 int32_t WebRtcSpl_DivW32HiLow(int32_t num, int16_t den_hi, int16_t den_low);
580 // End: Divisions.
581 
582 int32_t WebRtcSpl_Energy(int16_t* vector,
583                          size_t vector_length,
584                          int* scale_factor);
585 
586 // Filter operations.
587 size_t WebRtcSpl_FilterAR(const int16_t* ar_coef,
588                           size_t ar_coef_length,
589                           const int16_t* in_vector,
590                           size_t in_vector_length,
591                           int16_t* filter_state,
592                           size_t filter_state_length,
593                           int16_t* filter_state_low,
594                           size_t filter_state_low_length,
595                           int16_t* out_vector,
596                           int16_t* out_vector_low,
597                           size_t out_vector_low_length);
598 
599 // WebRtcSpl_FilterMAFastQ12(...)
600 //
601 // Performs a MA filtering on a vector in Q12
602 //
603 // Input:
604 //      - in_vector         : Input samples (state in positions
605 //                            in_vector[-order] .. in_vector[-1])
606 //      - ma_coef           : Filter coefficients (in Q12)
607 //      - ma_coef_length    : Number of B coefficients (order+1)
608 //      - vector_length     : Number of samples to be filtered
609 //
610 // Output:
611 //      - out_vector        : Filtered samples
612 //
613 void WebRtcSpl_FilterMAFastQ12(const int16_t* in_vector,
614                                int16_t* out_vector,
615                                const int16_t* ma_coef,
616                                size_t ma_coef_length,
617                                size_t vector_length);
618 
619 // Performs a AR filtering on a vector in Q12
620 // Input:
621 //      - data_in            : Input samples
622 //      - data_out           : State information in positions
623 //                               data_out[-order] .. data_out[-1]
624 //      - coefficients       : Filter coefficients (in Q12)
625 //      - coefficients_length: Number of coefficients (order+1)
626 //      - data_length        : Number of samples to be filtered
627 // Output:
628 //      - data_out           : Filtered samples
629 void WebRtcSpl_FilterARFastQ12(const int16_t* data_in,
630                                int16_t* data_out,
631                                const int16_t* __restrict coefficients,
632                                size_t coefficients_length,
633                                size_t data_length);
634 
635 // The functions (with related pointer) perform a MA down sampling filter
636 // on a vector.
637 // Input:
638 //      - data_in            : Input samples (state in positions
639 //                               data_in[-order] .. data_in[-1])
640 //      - data_in_length     : Number of samples in |data_in| to be filtered.
641 //                               This must be at least
642 //                               |delay| + |factor|*(|out_vector_length|-1) + 1)
643 //      - data_out_length    : Number of down sampled samples desired
644 //      - coefficients       : Filter coefficients (in Q12)
645 //      - coefficients_length: Number of coefficients (order+1)
646 //      - factor             : Decimation factor
647 //      - delay              : Delay of filter (compensated for in out_vector)
648 // Output:
649 //      - data_out           : Filtered samples
650 // Return value              : 0 if OK, -1 if |in_vector| is too short
651 typedef int (*DownsampleFast)(const int16_t* data_in,
652                               size_t data_in_length,
653                               int16_t* data_out,
654                               size_t data_out_length,
655                               const int16_t* __restrict coefficients,
656                               size_t coefficients_length,
657                               int factor,
658                               size_t delay);
659 extern const DownsampleFast WebRtcSpl_DownsampleFast;
660 int WebRtcSpl_DownsampleFastC(const int16_t* data_in,
661                               size_t data_in_length,
662                               int16_t* data_out,
663                               size_t data_out_length,
664                               const int16_t* __restrict coefficients,
665                               size_t coefficients_length,
666                               int factor,
667                               size_t delay);
668 #if defined(WEBRTC_HAS_NEON)
669 int WebRtcSpl_DownsampleFastNeon(const int16_t* data_in,
670                                  size_t data_in_length,
671                                  int16_t* data_out,
672                                  size_t data_out_length,
673                                  const int16_t* __restrict coefficients,
674                                  size_t coefficients_length,
675                                  int factor,
676                                  size_t delay);
677 #endif
678 #if defined(MIPS32_LE)
679 int WebRtcSpl_DownsampleFast_mips(const int16_t* data_in,
680                                   size_t data_in_length,
681                                   int16_t* data_out,
682                                   size_t data_out_length,
683                                   const int16_t* __restrict coefficients,
684                                   size_t coefficients_length,
685                                   int factor,
686                                   size_t delay);
687 #endif
688 
689 // End: Filter operations.
690 
691 // FFT operations
692 
693 int WebRtcSpl_ComplexFFT(int16_t vector[], int stages, int mode);
694 int WebRtcSpl_ComplexIFFT(int16_t vector[], int stages, int mode);
695 
696 // Treat a 16-bit complex data buffer |complex_data| as an array of 32-bit
697 // values, and swap elements whose indexes are bit-reverses of each other.
698 //
699 // Input:
700 //      - complex_data  : Complex data buffer containing 2^|stages| real
701 //                        elements interleaved with 2^|stages| imaginary
702 //                        elements: [Re Im Re Im Re Im....]
703 //      - stages        : Number of FFT stages. Must be at least 3 and at most
704 //                        10, since the table WebRtcSpl_kSinTable1024[] is 1024
705 //                        elements long.
706 //
707 // Output:
708 //      - complex_data  : The complex data buffer.
709 
710 void WebRtcSpl_ComplexBitReverse(int16_t* __restrict complex_data, int stages);
711 
712 // End: FFT operations
713 
714 /************************************************************
715  *
716  * RESAMPLING FUNCTIONS AND THEIR STRUCTS ARE DEFINED BELOW
717  *
718  ************************************************************/
719 
720 /*******************************************************************
721  * resample.c
722  *
723  * Includes the following resampling combinations
724  * 22 kHz -> 16 kHz
725  * 16 kHz -> 22 kHz
726  * 22 kHz ->  8 kHz
727  *  8 kHz -> 22 kHz
728  *
729  ******************************************************************/
730 
731 // state structure for 22 -> 16 resampler
732 typedef struct {
733   int32_t S_22_44[8];
734   int32_t S_44_32[8];
735   int32_t S_32_16[8];
736 } WebRtcSpl_State22khzTo16khz;
737 
738 void WebRtcSpl_Resample22khzTo16khz(const int16_t* in,
739                                     int16_t* out,
740                                     WebRtcSpl_State22khzTo16khz* state,
741                                     int32_t* tmpmem);
742 
743 void WebRtcSpl_ResetResample22khzTo16khz(WebRtcSpl_State22khzTo16khz* state);
744 
745 // state structure for 16 -> 22 resampler
746 typedef struct {
747   int32_t S_16_32[8];
748   int32_t S_32_22[8];
749 } WebRtcSpl_State16khzTo22khz;
750 
751 void WebRtcSpl_Resample16khzTo22khz(const int16_t* in,
752                                     int16_t* out,
753                                     WebRtcSpl_State16khzTo22khz* state,
754                                     int32_t* tmpmem);
755 
756 void WebRtcSpl_ResetResample16khzTo22khz(WebRtcSpl_State16khzTo22khz* state);
757 
758 // state structure for 22 -> 8 resampler
759 typedef struct {
760   int32_t S_22_22[16];
761   int32_t S_22_16[8];
762   int32_t S_16_8[8];
763 } WebRtcSpl_State22khzTo8khz;
764 
765 void WebRtcSpl_Resample22khzTo8khz(const int16_t* in,
766                                    int16_t* out,
767                                    WebRtcSpl_State22khzTo8khz* state,
768                                    int32_t* tmpmem);
769 
770 void WebRtcSpl_ResetResample22khzTo8khz(WebRtcSpl_State22khzTo8khz* state);
771 
772 // state structure for 8 -> 22 resampler
773 typedef struct {
774   int32_t S_8_16[8];
775   int32_t S_16_11[8];
776   int32_t S_11_22[8];
777 } WebRtcSpl_State8khzTo22khz;
778 
779 void WebRtcSpl_Resample8khzTo22khz(const int16_t* in,
780                                    int16_t* out,
781                                    WebRtcSpl_State8khzTo22khz* state,
782                                    int32_t* tmpmem);
783 
784 void WebRtcSpl_ResetResample8khzTo22khz(WebRtcSpl_State8khzTo22khz* state);
785 
786 /*******************************************************************
787  * resample_fractional.c
788  * Functions for internal use in the other resample functions
789  *
790  * Includes the following resampling combinations
791  * 48 kHz -> 32 kHz
792  * 32 kHz -> 24 kHz
793  * 44 kHz -> 32 kHz
794  *
795  ******************************************************************/
796 
797 void WebRtcSpl_Resample48khzTo32khz(const int32_t* In, int32_t* Out, size_t K);
798 
799 void WebRtcSpl_Resample32khzTo24khz(const int32_t* In, int32_t* Out, size_t K);
800 
801 void WebRtcSpl_Resample44khzTo32khz(const int32_t* In, int32_t* Out, size_t K);
802 
803 /*******************************************************************
804  * resample_48khz.c
805  *
806  * Includes the following resampling combinations
807  * 48 kHz -> 16 kHz
808  * 16 kHz -> 48 kHz
809  * 48 kHz ->  8 kHz
810  *  8 kHz -> 48 kHz
811  *
812  ******************************************************************/
813 
814 typedef struct {
815   int32_t S_48_48[16];
816   int32_t S_48_32[8];
817   int32_t S_32_16[8];
818 } WebRtcSpl_State48khzTo16khz;
819 
820 void WebRtcSpl_Resample48khzTo16khz(const int16_t* in,
821                                     int16_t* out,
822                                     WebRtcSpl_State48khzTo16khz* state,
823                                     int32_t* tmpmem);
824 
825 void WebRtcSpl_ResetResample48khzTo16khz(WebRtcSpl_State48khzTo16khz* state);
826 
827 typedef struct {
828   int32_t S_16_32[8];
829   int32_t S_32_24[8];
830   int32_t S_24_48[8];
831 } WebRtcSpl_State16khzTo48khz;
832 
833 void WebRtcSpl_Resample16khzTo48khz(const int16_t* in,
834                                     int16_t* out,
835                                     WebRtcSpl_State16khzTo48khz* state,
836                                     int32_t* tmpmem);
837 
838 void WebRtcSpl_ResetResample16khzTo48khz(WebRtcSpl_State16khzTo48khz* state);
839 
840 typedef struct {
841   int32_t S_48_24[8];
842   int32_t S_24_24[16];
843   int32_t S_24_16[8];
844   int32_t S_16_8[8];
845 } WebRtcSpl_State48khzTo8khz;
846 
847 void WebRtcSpl_Resample48khzTo8khz(const int16_t* in,
848                                    int16_t* out,
849                                    WebRtcSpl_State48khzTo8khz* state,
850                                    int32_t* tmpmem);
851 
852 void WebRtcSpl_ResetResample48khzTo8khz(WebRtcSpl_State48khzTo8khz* state);
853 
854 typedef struct {
855   int32_t S_8_16[8];
856   int32_t S_16_12[8];
857   int32_t S_12_24[8];
858   int32_t S_24_48[8];
859 } WebRtcSpl_State8khzTo48khz;
860 
861 void WebRtcSpl_Resample8khzTo48khz(const int16_t* in,
862                                    int16_t* out,
863                                    WebRtcSpl_State8khzTo48khz* state,
864                                    int32_t* tmpmem);
865 
866 void WebRtcSpl_ResetResample8khzTo48khz(WebRtcSpl_State8khzTo48khz* state);
867 
868 /*******************************************************************
869  * resample_by_2.c
870  *
871  * Includes down and up sampling by a factor of two.
872  *
873  ******************************************************************/
874 
875 void WebRtcSpl_DownsampleBy2(const int16_t* in,
876                              size_t len,
877                              int16_t* out,
878                              int32_t* filtState);
879 
880 void WebRtcSpl_UpsampleBy2(const int16_t* in,
881                            size_t len,
882                            int16_t* out,
883                            int32_t* filtState);
884 
885 /************************************************************
886  * END OF RESAMPLING FUNCTIONS
887  ************************************************************/
888 void WebRtcSpl_AnalysisQMF(const int16_t* in_data,
889                            size_t in_data_length,
890                            int16_t* low_band,
891                            int16_t* high_band,
892                            int32_t* filter_state1,
893                            int32_t* filter_state2);
894 void WebRtcSpl_SynthesisQMF(const int16_t* low_band,
895                             const int16_t* high_band,
896                             size_t band_length,
897                             int16_t* out_data,
898                             int32_t* filter_state1,
899                             int32_t* filter_state2);
900 
901 #ifdef __cplusplus
902 }
903 #endif  // __cplusplus
904 #endif  // COMMON_AUDIO_SIGNAL_PROCESSING_INCLUDE_SIGNAL_PROCESSING_LIBRARY_H_
905 
906 //
907 // WebRtcSpl_AddSatW16(...)
908 // WebRtcSpl_AddSatW32(...)
909 //
910 // Returns the result of a saturated 16-bit, respectively 32-bit, addition of
911 // the numbers specified by the |var1| and |var2| parameters.
912 //
913 // Input:
914 //      - var1      : Input variable 1
915 //      - var2      : Input variable 2
916 //
917 // Return value     : Added and saturated value
918 //
919 
920 //
921 // WebRtcSpl_SubSatW16(...)
922 // WebRtcSpl_SubSatW32(...)
923 //
924 // Returns the result of a saturated 16-bit, respectively 32-bit, subtraction
925 // of the numbers specified by the |var1| and |var2| parameters.
926 //
927 // Input:
928 //      - var1      : Input variable 1
929 //      - var2      : Input variable 2
930 //
931 // Returned value   : Subtracted and saturated value
932 //
933 
934 //
935 // WebRtcSpl_GetSizeInBits(...)
936 //
937 // Returns the # of bits that are needed at the most to represent the number
938 // specified by the |value| parameter.
939 //
940 // Input:
941 //      - value     : Input value
942 //
943 // Return value     : Number of bits needed to represent |value|
944 //
945 
946 //
947 // WebRtcSpl_NormW32(...)
948 //
949 // Norm returns the # of left shifts required to 32-bit normalize the 32-bit
950 // signed number specified by the |value| parameter.
951 //
952 // Input:
953 //      - value     : Input value
954 //
955 // Return value     : Number of bit shifts needed to 32-bit normalize |value|
956 //
957 
958 //
959 // WebRtcSpl_NormW16(...)
960 //
961 // Norm returns the # of left shifts required to 16-bit normalize the 16-bit
962 // signed number specified by the |value| parameter.
963 //
964 // Input:
965 //      - value     : Input value
966 //
967 // Return value     : Number of bit shifts needed to 32-bit normalize |value|
968 //
969 
970 //
971 // WebRtcSpl_NormU32(...)
972 //
973 // Norm returns the # of left shifts required to 32-bit normalize the unsigned
974 // 32-bit number specified by the |value| parameter.
975 //
976 // Input:
977 //      - value     : Input value
978 //
979 // Return value     : Number of bit shifts needed to 32-bit normalize |value|
980 //
981 
982 //
983 // WebRtcSpl_GetScalingSquare(...)
984 //
985 // Returns the # of bits required to scale the samples specified in the
986 // |in_vector| parameter so that, if the squares of the samples are added the
987 // # of times specified by the |times| parameter, the 32-bit addition will not
988 // overflow (result in int32_t).
989 //
990 // Input:
991 //      - in_vector         : Input vector to check scaling on
992 //      - in_vector_length  : Samples in |in_vector|
993 //      - times             : Number of additions to be performed
994 //
995 // Return value             : Number of right bit shifts needed to avoid
996 //                            overflow in the addition calculation
997 //
998 
999 //
1000 // WebRtcSpl_MemSetW16(...)
1001 //
1002 // Sets all the values in the int16_t vector |vector| of length
1003 // |vector_length| to the specified value |set_value|
1004 //
1005 // Input:
1006 //      - vector        : Pointer to the int16_t vector
1007 //      - set_value     : Value specified
1008 //      - vector_length : Length of vector
1009 //
1010 
1011 //
1012 // WebRtcSpl_MemSetW32(...)
1013 //
1014 // Sets all the values in the int32_t vector |vector| of length
1015 // |vector_length| to the specified value |set_value|
1016 //
1017 // Input:
1018 //      - vector        : Pointer to the int16_t vector
1019 //      - set_value     : Value specified
1020 //      - vector_length : Length of vector
1021 //
1022 
1023 //
1024 // WebRtcSpl_MemCpyReversedOrder(...)
1025 //
1026 // Copies all the values from the source int16_t vector |in_vector| to a
1027 // destination int16_t vector |out_vector|. It is done in reversed order,
1028 // meaning that the first sample of |in_vector| is copied to the last sample of
1029 // the |out_vector|. The procedure continues until the last sample of
1030 // |in_vector| has been copied to the first sample of |out_vector|. This
1031 // creates a reversed vector. Used in e.g. prediction in iLBC.
1032 //
1033 // Input:
1034 //      - in_vector     : Pointer to the first sample in a int16_t vector
1035 //                        of length |length|
1036 //      - vector_length : Number of elements to copy
1037 //
1038 // Output:
1039 //      - out_vector    : Pointer to the last sample in a int16_t vector
1040 //                        of length |length|
1041 //
1042 
1043 //
1044 // WebRtcSpl_CopyFromEndW16(...)
1045 //
1046 // Copies the rightmost |samples| of |in_vector| (of length |in_vector_length|)
1047 // to the vector |out_vector|.
1048 //
1049 // Input:
1050 //      - in_vector         : Input vector
1051 //      - in_vector_length  : Number of samples in |in_vector|
1052 //      - samples           : Number of samples to extract (from right side)
1053 //                            from |in_vector|
1054 //
1055 // Output:
1056 //      - out_vector        : Vector with the requested samples
1057 //
1058 
1059 //
1060 // WebRtcSpl_ZerosArrayW16(...)
1061 // WebRtcSpl_ZerosArrayW32(...)
1062 //
1063 // Inserts the value "zero" in all positions of a w16 and a w32 vector
1064 // respectively.
1065 //
1066 // Input:
1067 //      - vector_length : Number of samples in vector
1068 //
1069 // Output:
1070 //      - vector        : Vector containing all zeros
1071 //
1072 
1073 //
1074 // WebRtcSpl_VectorBitShiftW16(...)
1075 // WebRtcSpl_VectorBitShiftW32(...)
1076 //
1077 // Bit shifts all the values in a vector up or downwards. Different calls for
1078 // int16_t and int32_t vectors respectively.
1079 //
1080 // Input:
1081 //      - vector_length : Length of vector
1082 //      - in_vector     : Pointer to the vector that should be bit shifted
1083 //      - right_shifts  : Number of right bit shifts (negative value gives left
1084 //                        shifts)
1085 //
1086 // Output:
1087 //      - out_vector    : Pointer to the result vector (can be the same as
1088 //                        |in_vector|)
1089 //
1090 
1091 //
1092 // WebRtcSpl_VectorBitShiftW32ToW16(...)
1093 //
1094 // Bit shifts all the values in a int32_t vector up or downwards and
1095 // stores the result as an int16_t vector. The function will saturate the
1096 // signal if needed, before storing in the output vector.
1097 //
1098 // Input:
1099 //      - vector_length : Length of vector
1100 //      - in_vector     : Pointer to the vector that should be bit shifted
1101 //      - right_shifts  : Number of right bit shifts (negative value gives left
1102 //                        shifts)
1103 //
1104 // Output:
1105 //      - out_vector    : Pointer to the result vector (can be the same as
1106 //                        |in_vector|)
1107 //
1108 
1109 //
1110 // WebRtcSpl_ScaleVector(...)
1111 //
1112 // Performs the vector operation:
1113 //  out_vector[k] = (gain*in_vector[k])>>right_shifts
1114 //
1115 // Input:
1116 //      - in_vector     : Input vector
1117 //      - gain          : Scaling gain
1118 //      - vector_length : Elements in the |in_vector|
1119 //      - right_shifts  : Number of right bit shifts applied
1120 //
1121 // Output:
1122 //      - out_vector    : Output vector (can be the same as |in_vector|)
1123 //
1124 
1125 //
1126 // WebRtcSpl_ScaleVectorWithSat(...)
1127 //
1128 // Performs the vector operation:
1129 //  out_vector[k] = SATURATE( (gain*in_vector[k])>>right_shifts )
1130 //
1131 // Input:
1132 //      - in_vector     : Input vector
1133 //      - gain          : Scaling gain
1134 //      - vector_length : Elements in the |in_vector|
1135 //      - right_shifts  : Number of right bit shifts applied
1136 //
1137 // Output:
1138 //      - out_vector    : Output vector (can be the same as |in_vector|)
1139 //
1140 
1141 //
1142 // WebRtcSpl_ScaleAndAddVectors(...)
1143 //
1144 // Performs the vector operation:
1145 //  out_vector[k] = (gain1*in_vector1[k])>>right_shifts1
1146 //                  + (gain2*in_vector2[k])>>right_shifts2
1147 //
1148 // Input:
1149 //      - in_vector1    : Input vector 1
1150 //      - gain1         : Gain to be used for vector 1
1151 //      - right_shifts1 : Right bit shift to be used for vector 1
1152 //      - in_vector2    : Input vector 2
1153 //      - gain2         : Gain to be used for vector 2
1154 //      - right_shifts2 : Right bit shift to be used for vector 2
1155 //      - vector_length : Elements in the input vectors
1156 //
1157 // Output:
1158 //      - out_vector    : Output vector
1159 //
1160 
1161 //
1162 // WebRtcSpl_ReverseOrderMultArrayElements(...)
1163 //
1164 // Performs the vector operation:
1165 //  out_vector[n] = (in_vector[n]*window[-n])>>right_shifts
1166 //
1167 // Input:
1168 //      - in_vector     : Input vector
1169 //      - window        : Window vector (should be reversed). The pointer
1170 //                        should be set to the last value in the vector
1171 //      - right_shifts  : Number of right bit shift to be applied after the
1172 //                        multiplication
1173 //      - vector_length : Number of elements in |in_vector|
1174 //
1175 // Output:
1176 //      - out_vector    : Output vector (can be same as |in_vector|)
1177 //
1178 
1179 //
1180 // WebRtcSpl_ElementwiseVectorMult(...)
1181 //
1182 // Performs the vector operation:
1183 //  out_vector[n] = (in_vector[n]*window[n])>>right_shifts
1184 //
1185 // Input:
1186 //      - in_vector     : Input vector
1187 //      - window        : Window vector.
1188 //      - right_shifts  : Number of right bit shift to be applied after the
1189 //                        multiplication
1190 //      - vector_length : Number of elements in |in_vector|
1191 //
1192 // Output:
1193 //      - out_vector    : Output vector (can be same as |in_vector|)
1194 //
1195 
1196 //
1197 // WebRtcSpl_AddVectorsAndShift(...)
1198 //
1199 // Performs the vector operation:
1200 //  out_vector[k] = (in_vector1[k] + in_vector2[k])>>right_shifts
1201 //
1202 // Input:
1203 //      - in_vector1    : Input vector 1
1204 //      - in_vector2    : Input vector 2
1205 //      - right_shifts  : Number of right bit shift to be applied after the
1206 //                        multiplication
1207 //      - vector_length : Number of elements in |in_vector1| and |in_vector2|
1208 //
1209 // Output:
1210 //      - out_vector    : Output vector (can be same as |in_vector1|)
1211 //
1212 
1213 //
1214 // WebRtcSpl_AddAffineVectorToVector(...)
1215 //
1216 // Adds an affine transformed vector to another vector |out_vector|, i.e,
1217 // performs
1218 //  out_vector[k] += (in_vector[k]*gain+add_constant)>>right_shifts
1219 //
1220 // Input:
1221 //      - in_vector     : Input vector
1222 //      - gain          : Gain value, used to multiply the in vector with
1223 //      - add_constant  : Constant value to add (usually 1<<(right_shifts-1),
1224 //                        but others can be used as well
1225 //      - right_shifts  : Number of right bit shifts (0-16)
1226 //      - vector_length : Number of samples in |in_vector| and |out_vector|
1227 //
1228 // Output:
1229 //      - out_vector    : Vector with the output
1230 //
1231 
1232 //
1233 // WebRtcSpl_AffineTransformVector(...)
1234 //
1235 // Affine transforms a vector, i.e, performs
1236 //  out_vector[k] = (in_vector[k]*gain+add_constant)>>right_shifts
1237 //
1238 // Input:
1239 //      - in_vector     : Input vector
1240 //      - gain          : Gain value, used to multiply the in vector with
1241 //      - add_constant  : Constant value to add (usually 1<<(right_shifts-1),
1242 //                        but others can be used as well
1243 //      - right_shifts  : Number of right bit shifts (0-16)
1244 //      - vector_length : Number of samples in |in_vector| and |out_vector|
1245 //
1246 // Output:
1247 //      - out_vector    : Vector with the output
1248 //
1249 
1250 //
1251 // WebRtcSpl_IncreaseSeed(...)
1252 //
1253 // Increases the seed (and returns the new value)
1254 //
1255 // Input:
1256 //      - seed      : Seed for random calculation
1257 //
1258 // Output:
1259 //      - seed      : Updated seed value
1260 //
1261 // Return value     : The new seed value
1262 //
1263 
1264 //
1265 // WebRtcSpl_RandU(...)
1266 //
1267 // Produces a uniformly distributed value in the int16_t range
1268 //
1269 // Input:
1270 //      - seed      : Seed for random calculation
1271 //
1272 // Output:
1273 //      - seed      : Updated seed value
1274 //
1275 // Return value     : Uniformly distributed value in the range
1276 //                    [Word16_MIN...Word16_MAX]
1277 //
1278 
1279 //
1280 // WebRtcSpl_RandN(...)
1281 //
1282 // Produces a normal distributed value in the int16_t range
1283 //
1284 // Input:
1285 //      - seed      : Seed for random calculation
1286 //
1287 // Output:
1288 //      - seed      : Updated seed value
1289 //
1290 // Return value     : N(0,1) value in the Q13 domain
1291 //
1292 
1293 //
1294 // WebRtcSpl_RandUArray(...)
1295 //
1296 // Produces a uniformly distributed vector with elements in the int16_t
1297 // range
1298 //
1299 // Input:
1300 //      - vector_length : Samples wanted in the vector
1301 //      - seed          : Seed for random calculation
1302 //
1303 // Output:
1304 //      - vector        : Vector with the uniform values
1305 //      - seed          : Updated seed value
1306 //
1307 // Return value         : Number of samples in vector, i.e., |vector_length|
1308 //
1309 
1310 //
1311 // WebRtcSpl_Sqrt(...)
1312 //
1313 // Returns the square root of the input value |value|. The precision of this
1314 // function is integer precision, i.e., sqrt(8) gives 2 as answer.
1315 // If |value| is a negative number then 0 is returned.
1316 //
1317 // Algorithm:
1318 //
1319 // A sixth order Taylor Series expansion is used here to compute the square
1320 // root of a number y^0.5 = (1+x)^0.5
1321 // where
1322 // x = y-1
1323 //   = 1+(x/2)-0.5*((x/2)^2+0.5*((x/2)^3-0.625*((x/2)^4+0.875*((x/2)^5)
1324 // 0.5 <= x < 1
1325 //
1326 // Input:
1327 //      - value     : Value to calculate sqrt of
1328 //
1329 // Return value     : Result of the sqrt calculation
1330 //
1331 
1332 //
1333 // WebRtcSpl_DivU32U16(...)
1334 //
1335 // Divides a uint32_t |num| by a uint16_t |den|.
1336 //
1337 // If |den|==0, (uint32_t)0xFFFFFFFF is returned.
1338 //
1339 // Input:
1340 //      - num       : Numerator
1341 //      - den       : Denominator
1342 //
1343 // Return value     : Result of the division (as a uint32_t), i.e., the
1344 //                    integer part of num/den.
1345 //
1346 
1347 //
1348 // WebRtcSpl_DivW32W16(...)
1349 //
1350 // Divides a int32_t |num| by a int16_t |den|.
1351 //
1352 // If |den|==0, (int32_t)0x7FFFFFFF is returned.
1353 //
1354 // Input:
1355 //      - num       : Numerator
1356 //      - den       : Denominator
1357 //
1358 // Return value     : Result of the division (as a int32_t), i.e., the
1359 //                    integer part of num/den.
1360 //
1361 
1362 //
1363 // WebRtcSpl_DivW32W16ResW16(...)
1364 //
1365 // Divides a int32_t |num| by a int16_t |den|, assuming that the
1366 // result is less than 32768, otherwise an unpredictable result will occur.
1367 //
1368 // If |den|==0, (int16_t)0x7FFF is returned.
1369 //
1370 // Input:
1371 //      - num       : Numerator
1372 //      - den       : Denominator
1373 //
1374 // Return value     : Result of the division (as a int16_t), i.e., the
1375 //                    integer part of num/den.
1376 //
1377 
1378 //
1379 // WebRtcSpl_DivResultInQ31(...)
1380 //
1381 // Divides a int32_t |num| by a int16_t |den|, assuming that the
1382 // absolute value of the denominator is larger than the numerator, otherwise
1383 // an unpredictable result will occur.
1384 //
1385 // Input:
1386 //      - num       : Numerator
1387 //      - den       : Denominator
1388 //
1389 // Return value     : Result of the division in Q31.
1390 //
1391 
1392 //
1393 // WebRtcSpl_DivW32HiLow(...)
1394 //
1395 // Divides a int32_t |num| by a denominator in hi, low format. The
1396 // absolute value of the denominator has to be larger (or equal to) the
1397 // numerator.
1398 //
1399 // Input:
1400 //      - num       : Numerator
1401 //      - den_hi    : High part of denominator
1402 //      - den_low   : Low part of denominator
1403 //
1404 // Return value     : Divided value in Q31
1405 //
1406 
1407 //
1408 // WebRtcSpl_Energy(...)
1409 //
1410 // Calculates the energy of a vector
1411 //
1412 // Input:
1413 //      - vector        : Vector which the energy should be calculated on
1414 //      - vector_length : Number of samples in vector
1415 //
1416 // Output:
1417 //      - scale_factor  : Number of left bit shifts needed to get the physical
1418 //                        energy value, i.e, to get the Q0 value
1419 //
1420 // Return value         : Energy value in Q(-|scale_factor|)
1421 //
1422 
1423 //
1424 // WebRtcSpl_FilterAR(...)
1425 //
1426 // Performs a 32-bit AR filtering on a vector in Q12
1427 //
1428 // Input:
1429 //  - ar_coef                   : AR-coefficient vector (values in Q12),
1430 //                                ar_coef[0] must be 4096.
1431 //  - ar_coef_length            : Number of coefficients in |ar_coef|.
1432 //  - in_vector                 : Vector to be filtered.
1433 //  - in_vector_length          : Number of samples in |in_vector|.
1434 //  - filter_state              : Current state (higher part) of the filter.
1435 //  - filter_state_length       : Length (in samples) of |filter_state|.
1436 //  - filter_state_low          : Current state (lower part) of the filter.
1437 //  - filter_state_low_length   : Length (in samples) of |filter_state_low|.
1438 //  - out_vector_low_length     : Maximum length (in samples) of
1439 //                                |out_vector_low|.
1440 //
1441 // Output:
1442 //  - filter_state              : Updated state (upper part) vector.
1443 //  - filter_state_low          : Updated state (lower part) vector.
1444 //  - out_vector                : Vector containing the upper part of the
1445 //                                filtered values.
1446 //  - out_vector_low            : Vector containing the lower part of the
1447 //                                filtered values.
1448 //
1449 // Return value                 : Number of samples in the |out_vector|.
1450 //
1451 
1452 //
1453 // WebRtcSpl_ComplexIFFT(...)
1454 //
1455 // Complex Inverse FFT
1456 //
1457 // Computes an inverse complex 2^|stages|-point FFT on the input vector, which
1458 // is in bit-reversed order. The original content of the vector is destroyed in
1459 // the process, since the input is overwritten by the output, normal-ordered,
1460 // FFT vector. With X as the input complex vector, y as the output complex
1461 // vector and with M = 2^|stages|, the following is computed:
1462 //
1463 //        M-1
1464 // y(k) = sum[X(i)*[cos(2*pi*i*k/M) + j*sin(2*pi*i*k/M)]]
1465 //        i=0
1466 //
1467 // The implementations are optimized for speed, not for code size. It uses the
1468 // decimation-in-time algorithm with radix-2 butterfly technique.
1469 //
1470 // Input:
1471 //      - vector    : In pointer to complex vector containing 2^|stages|
1472 //                    real elements interleaved with 2^|stages| imaginary
1473 //                    elements.
1474 //                    [ReImReImReIm....]
1475 //                    The elements are in Q(-scale) domain, see more on Return
1476 //                    Value below.
1477 //
1478 //      - stages    : Number of FFT stages. Must be at least 3 and at most 10,
1479 //                    since the table WebRtcSpl_kSinTable1024[] is 1024
1480 //                    elements long.
1481 //
1482 //      - mode      : This parameter gives the user to choose how the FFT
1483 //                    should work.
1484 //                    mode==0: Low-complexity and Low-accuracy mode
1485 //                    mode==1: High-complexity and High-accuracy mode
1486 //
1487 // Output:
1488 //      - vector    : Out pointer to the FFT vector (the same as input).
1489 //
1490 // Return Value     : The scale value that tells the number of left bit shifts
1491 //                    that the elements in the |vector| should be shifted with
1492 //                    in order to get Q0 values, i.e. the physically correct
1493 //                    values. The scale parameter is always 0 or positive,
1494 //                    except if N>1024 (|stages|>10), which returns a scale
1495 //                    value of -1, indicating error.
1496 //
1497 
1498 //
1499 // WebRtcSpl_ComplexFFT(...)
1500 //
1501 // Complex FFT
1502 //
1503 // Computes a complex 2^|stages|-point FFT on the input vector, which is in
1504 // bit-reversed order. The original content of the vector is destroyed in
1505 // the process, since the input is overwritten by the output, normal-ordered,
1506 // FFT vector. With x as the input complex vector, Y as the output complex
1507 // vector and with M = 2^|stages|, the following is computed:
1508 //
1509 //              M-1
1510 // Y(k) = 1/M * sum[x(i)*[cos(2*pi*i*k/M) + j*sin(2*pi*i*k/M)]]
1511 //              i=0
1512 //
1513 // The implementations are optimized for speed, not for code size. It uses the
1514 // decimation-in-time algorithm with radix-2 butterfly technique.
1515 //
1516 // This routine prevents overflow by scaling by 2 before each FFT stage. This is
1517 // a fixed scaling, for proper normalization - there will be log2(n) passes, so
1518 // this results in an overall factor of 1/n, distributed to maximize arithmetic
1519 // accuracy.
1520 //
1521 // Input:
1522 //      - vector    : In pointer to complex vector containing 2^|stages| real
1523 //                    elements interleaved with 2^|stages| imaginary elements.
1524 //                    [ReImReImReIm....]
1525 //                    The output is in the Q0 domain.
1526 //
1527 //      - stages    : Number of FFT stages. Must be at least 3 and at most 10,
1528 //                    since the table WebRtcSpl_kSinTable1024[] is 1024
1529 //                    elements long.
1530 //
1531 //      - mode      : This parameter gives the user to choose how the FFT
1532 //                    should work.
1533 //                    mode==0: Low-complexity and Low-accuracy mode
1534 //                    mode==1: High-complexity and High-accuracy mode
1535 //
1536 // Output:
1537 //      - vector    : The output FFT vector is in the Q0 domain.
1538 //
1539 // Return value     : The scale parameter is always 0, except if N>1024,
1540 //                    which returns a scale value of -1, indicating error.
1541 //
1542 
1543 //
1544 // WebRtcSpl_AnalysisQMF(...)
1545 //
1546 // Splits a 0-2*F Hz signal into two sub bands: 0-F Hz and F-2*F Hz. The
1547 // current version has F = 8000, therefore, a super-wideband audio signal is
1548 // split to lower-band 0-8 kHz and upper-band 8-16 kHz.
1549 //
1550 // Input:
1551 //      - in_data       : Wide band speech signal, 320 samples (10 ms)
1552 //
1553 // Input & Output:
1554 //      - filter_state1 : Filter state for first All-pass filter
1555 //      - filter_state2 : Filter state for second All-pass filter
1556 //
1557 // Output:
1558 //      - low_band      : Lower-band signal 0-8 kHz band, 160 samples (10 ms)
1559 //      - high_band     : Upper-band signal 8-16 kHz band (flipped in frequency
1560 //                        domain), 160 samples (10 ms)
1561 //
1562 
1563 //
1564 // WebRtcSpl_SynthesisQMF(...)
1565 //
1566 // Combines the two sub bands (0-F and F-2*F Hz) into a signal of 0-2*F
1567 // Hz, (current version has F = 8000 Hz). So the filter combines lower-band
1568 // (0-8 kHz) and upper-band (8-16 kHz) channels to obtain super-wideband 0-16
1569 // kHz audio.
1570 //
1571 // Input:
1572 //      - low_band      : The signal with the 0-8 kHz band, 160 samples (10 ms)
1573 //      - high_band     : The signal with the 8-16 kHz band, 160 samples (10 ms)
1574 //
1575 // Input & Output:
1576 //      - filter_state1 : Filter state for first All-pass filter
1577 //      - filter_state2 : Filter state for second All-pass filter
1578 //
1579 // Output:
1580 //      - out_data      : Super-wideband speech signal, 0-16 kHz
1581 //
1582 
1583 // int16_t WebRtcSpl_SatW32ToW16(...)
1584 //
1585 // This function saturates a 32-bit word into a 16-bit word.
1586 //
1587 // Input:
1588 //      - value32   : The value of a 32-bit word.
1589 //
1590 // Output:
1591 //      - out16     : the saturated 16-bit word.
1592 //
1593 
1594 // int32_t WebRtc_MulAccumW16(...)
1595 //
1596 // This function multiply a 16-bit word by a 16-bit word, and accumulate this
1597 // value to a 32-bit integer.
1598 //
1599 // Input:
1600 //      - a    : The value of the first 16-bit word.
1601 //      - b    : The value of the second 16-bit word.
1602 //      - c    : The value of an 32-bit integer.
1603 //
1604 // Return Value: The value of a * b + c.
1605 //
1606