]> git.gir.st - tmk_keyboard.git/blob - tmk_core/tool/mbed/mbed-sdk/libraries/dsp/cmsis_dsp/FilteringFunctions/arm_fir_sparse_q15.c
Merge commit '71381457fa1311dfa0b58ba882a96db740640871'
[tmk_keyboard.git] / tmk_core / tool / mbed / mbed-sdk / libraries / dsp / cmsis_dsp / FilteringFunctions / arm_fir_sparse_q15.c
1 /* ----------------------------------------------------------------------
2 * Copyright (C) 2010-2013 ARM Limited. All rights reserved.
3 *
4 * $Date: 17. January 2013
5 * $Revision: V1.4.1
6 *
7 * Project: CMSIS DSP Library
8 * Title: arm_fir_sparse_q15.c
9 *
10 * Description: Q15 sparse FIR filter processing function.
11 *
12 * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0
13 *
14 * Redistribution and use in source and binary forms, with or without
15 * modification, are permitted provided that the following conditions
16 * are met:
17 * - Redistributions of source code must retain the above copyright
18 * notice, this list of conditions and the following disclaimer.
19 * - Redistributions in binary form must reproduce the above copyright
20 * notice, this list of conditions and the following disclaimer in
21 * the documentation and/or other materials provided with the
22 * distribution.
23 * - Neither the name of ARM LIMITED nor the names of its contributors
24 * may be used to endorse or promote products derived from this
25 * software without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
28 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
29 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
30 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
31 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
32 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
33 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
34 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
35 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
36 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
37 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
38 * POSSIBILITY OF SUCH DAMAGE.
39 * ------------------------------------------------------------------- */
40 #include "arm_math.h"
41
42 /**
43 * @addtogroup FIR_Sparse
44 * @{
45 */
46
47 /**
48 * @brief Processing function for the Q15 sparse FIR filter.
49 * @param[in] *S points to an instance of the Q15 sparse FIR structure.
50 * @param[in] *pSrc points to the block of input data.
51 * @param[out] *pDst points to the block of output data
52 * @param[in] *pScratchIn points to a temporary buffer of size blockSize.
53 * @param[in] *pScratchOut points to a temporary buffer of size blockSize.
54 * @param[in] blockSize number of input samples to process per call.
55 * @return none.
56 *
57 * <b>Scaling and Overflow Behavior:</b>
58 * \par
59 * The function is implemented using an internal 32-bit accumulator.
60 * The 1.15 x 1.15 multiplications yield a 2.30 result and these are added to a 2.30 accumulator.
61 * Thus the full precision of the multiplications is maintained but there is only a single guard bit in the accumulator.
62 * If the accumulator result overflows it will wrap around rather than saturate.
63 * After all multiply-accumulates are performed, the 2.30 accumulator is truncated to 2.15 format and then saturated to 1.15 format.
64 * In order to avoid overflows the input signal or coefficients must be scaled down by log2(numTaps) bits.
65 */
66
67
68 void arm_fir_sparse_q15(
69 arm_fir_sparse_instance_q15 * S,
70 q15_t * pSrc,
71 q15_t * pDst,
72 q15_t * pScratchIn,
73 q31_t * pScratchOut,
74 uint32_t blockSize)
75 {
76
77 q15_t *pState = S->pState; /* State pointer */
78 q15_t *pIn = pSrc; /* Working pointer for input */
79 q15_t *pOut = pDst; /* Working pointer for output */
80 q15_t *pCoeffs = S->pCoeffs; /* Coefficient pointer */
81 q15_t *px; /* Temporary pointers for scratch buffer */
82 q15_t *pb = pScratchIn; /* Temporary pointers for scratch buffer */
83 q15_t *py = pState; /* Temporary pointers for state buffer */
84 int32_t *pTapDelay = S->pTapDelay; /* Pointer to the array containing offset of the non-zero tap values. */
85 uint32_t delaySize = S->maxDelay + blockSize; /* state length */
86 uint16_t numTaps = S->numTaps; /* Filter order */
87 int32_t readIndex; /* Read index of the state buffer */
88 uint32_t tapCnt, blkCnt; /* loop counters */
89 q15_t coeff = *pCoeffs++; /* Read the first coefficient value */
90 q31_t *pScr2 = pScratchOut; /* Working pointer for pScratchOut */
91
92
93 #ifndef ARM_MATH_CM0_FAMILY
94
95 /* Run the below code for Cortex-M4 and Cortex-M3 */
96
97 q31_t in1, in2; /* Temporary variables */
98
99
100 /* BlockSize of Input samples are copied into the state buffer */
101 /* StateIndex points to the starting position to write in the state buffer */
102 arm_circularWrite_q15(py, delaySize, &S->stateIndex, 1, pIn, 1, blockSize);
103
104 /* Loop over the number of taps. */
105 tapCnt = numTaps;
106
107 /* Read Index, from where the state buffer should be read, is calculated. */
108 readIndex = (S->stateIndex - blockSize) - *pTapDelay++;
109
110 /* Wraparound of readIndex */
111 if(readIndex < 0)
112 {
113 readIndex += (int32_t) delaySize;
114 }
115
116 /* Working pointer for state buffer is updated */
117 py = pState;
118
119 /* blockSize samples are read from the state buffer */
120 arm_circularRead_q15(py, delaySize, &readIndex, 1,
121 pb, pb, blockSize, 1, blockSize);
122
123 /* Working pointer for the scratch buffer of state values */
124 px = pb;
125
126 /* Working pointer for scratch buffer of output values */
127 pScratchOut = pScr2;
128
129 /* Loop over the blockSize. Unroll by a factor of 4.
130 * Compute 4 multiplications at a time. */
131 blkCnt = blockSize >> 2;
132
133 while(blkCnt > 0u)
134 {
135 /* Perform multiplication and store in the scratch buffer */
136 *pScratchOut++ = ((q31_t) * px++ * coeff);
137 *pScratchOut++ = ((q31_t) * px++ * coeff);
138 *pScratchOut++ = ((q31_t) * px++ * coeff);
139 *pScratchOut++ = ((q31_t) * px++ * coeff);
140
141 /* Decrement the loop counter */
142 blkCnt--;
143 }
144
145 /* If the blockSize is not a multiple of 4,
146 * compute the remaining samples */
147 blkCnt = blockSize % 0x4u;
148
149 while(blkCnt > 0u)
150 {
151 /* Perform multiplication and store in the scratch buffer */
152 *pScratchOut++ = ((q31_t) * px++ * coeff);
153
154 /* Decrement the loop counter */
155 blkCnt--;
156 }
157
158 /* Load the coefficient value and
159 * increment the coefficient buffer for the next set of state values */
160 coeff = *pCoeffs++;
161
162 /* Read Index, from where the state buffer should be read, is calculated. */
163 readIndex = (S->stateIndex - blockSize) - *pTapDelay++;
164
165 /* Wraparound of readIndex */
166 if(readIndex < 0)
167 {
168 readIndex += (int32_t) delaySize;
169 }
170
171 /* Loop over the number of taps. */
172 tapCnt = (uint32_t) numTaps - 1u;
173
174 while(tapCnt > 0u)
175 {
176 /* Working pointer for state buffer is updated */
177 py = pState;
178
179 /* blockSize samples are read from the state buffer */
180 arm_circularRead_q15(py, delaySize, &readIndex, 1,
181 pb, pb, blockSize, 1, blockSize);
182
183 /* Working pointer for the scratch buffer of state values */
184 px = pb;
185
186 /* Working pointer for scratch buffer of output values */
187 pScratchOut = pScr2;
188
189 /* Loop over the blockSize. Unroll by a factor of 4.
190 * Compute 4 MACS at a time. */
191 blkCnt = blockSize >> 2;
192
193 while(blkCnt > 0u)
194 {
195 /* Perform Multiply-Accumulate */
196 *pScratchOut++ += (q31_t) * px++ * coeff;
197 *pScratchOut++ += (q31_t) * px++ * coeff;
198 *pScratchOut++ += (q31_t) * px++ * coeff;
199 *pScratchOut++ += (q31_t) * px++ * coeff;
200
201 /* Decrement the loop counter */
202 blkCnt--;
203 }
204
205 /* If the blockSize is not a multiple of 4,
206 * compute the remaining samples */
207 blkCnt = blockSize % 0x4u;
208
209 while(blkCnt > 0u)
210 {
211 /* Perform Multiply-Accumulate */
212 *pScratchOut++ += (q31_t) * px++ * coeff;
213
214 /* Decrement the loop counter */
215 blkCnt--;
216 }
217
218 /* Load the coefficient value and
219 * increment the coefficient buffer for the next set of state values */
220 coeff = *pCoeffs++;
221
222 /* Read Index, from where the state buffer should be read, is calculated. */
223 readIndex = (S->stateIndex - blockSize) - *pTapDelay++;
224
225 /* Wraparound of readIndex */
226 if(readIndex < 0)
227 {
228 readIndex += (int32_t) delaySize;
229 }
230
231 /* Decrement the tap loop counter */
232 tapCnt--;
233 }
234
235 /* All the output values are in pScratchOut buffer.
236 Convert them into 1.15 format, saturate and store in the destination buffer. */
237 /* Loop over the blockSize. */
238 blkCnt = blockSize >> 2;
239
240 while(blkCnt > 0u)
241 {
242 in1 = *pScr2++;
243 in2 = *pScr2++;
244
245 #ifndef ARM_MATH_BIG_ENDIAN
246
247 *__SIMD32(pOut)++ =
248 __PKHBT((q15_t) __SSAT(in1 >> 15, 16), (q15_t) __SSAT(in2 >> 15, 16),
249 16);
250
251 #else
252 *__SIMD32(pOut)++ =
253 __PKHBT((q15_t) __SSAT(in2 >> 15, 16), (q15_t) __SSAT(in1 >> 15, 16),
254 16);
255
256 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
257
258 in1 = *pScr2++;
259
260 in2 = *pScr2++;
261
262 #ifndef ARM_MATH_BIG_ENDIAN
263
264 *__SIMD32(pOut)++ =
265 __PKHBT((q15_t) __SSAT(in1 >> 15, 16), (q15_t) __SSAT(in2 >> 15, 16),
266 16);
267
268 #else
269
270 *__SIMD32(pOut)++ =
271 __PKHBT((q15_t) __SSAT(in2 >> 15, 16), (q15_t) __SSAT(in1 >> 15, 16),
272 16);
273
274 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
275
276
277 blkCnt--;
278
279 }
280
281 /* If the blockSize is not a multiple of 4,
282 remaining samples are processed in the below loop */
283 blkCnt = blockSize % 0x4u;
284
285 while(blkCnt > 0u)
286 {
287 *pOut++ = (q15_t) __SSAT(*pScr2++ >> 15, 16);
288 blkCnt--;
289 }
290
291 #else
292
293 /* Run the below code for Cortex-M0 */
294
295 /* BlockSize of Input samples are copied into the state buffer */
296 /* StateIndex points to the starting position to write in the state buffer */
297 arm_circularWrite_q15(py, delaySize, &S->stateIndex, 1, pIn, 1, blockSize);
298
299 /* Loop over the number of taps. */
300 tapCnt = numTaps;
301
302 /* Read Index, from where the state buffer should be read, is calculated. */
303 readIndex = (S->stateIndex - blockSize) - *pTapDelay++;
304
305 /* Wraparound of readIndex */
306 if(readIndex < 0)
307 {
308 readIndex += (int32_t) delaySize;
309 }
310
311 /* Working pointer for state buffer is updated */
312 py = pState;
313
314 /* blockSize samples are read from the state buffer */
315 arm_circularRead_q15(py, delaySize, &readIndex, 1,
316 pb, pb, blockSize, 1, blockSize);
317
318 /* Working pointer for the scratch buffer of state values */
319 px = pb;
320
321 /* Working pointer for scratch buffer of output values */
322 pScratchOut = pScr2;
323
324 blkCnt = blockSize;
325
326 while(blkCnt > 0u)
327 {
328 /* Perform multiplication and store in the scratch buffer */
329 *pScratchOut++ = ((q31_t) * px++ * coeff);
330
331 /* Decrement the loop counter */
332 blkCnt--;
333 }
334
335 /* Load the coefficient value and
336 * increment the coefficient buffer for the next set of state values */
337 coeff = *pCoeffs++;
338
339 /* Read Index, from where the state buffer should be read, is calculated. */
340 readIndex = (S->stateIndex - blockSize) - *pTapDelay++;
341
342 /* Wraparound of readIndex */
343 if(readIndex < 0)
344 {
345 readIndex += (int32_t) delaySize;
346 }
347
348 /* Loop over the number of taps. */
349 tapCnt = (uint32_t) numTaps - 1u;
350
351 while(tapCnt > 0u)
352 {
353 /* Working pointer for state buffer is updated */
354 py = pState;
355
356 /* blockSize samples are read from the state buffer */
357 arm_circularRead_q15(py, delaySize, &readIndex, 1,
358 pb, pb, blockSize, 1, blockSize);
359
360 /* Working pointer for the scratch buffer of state values */
361 px = pb;
362
363 /* Working pointer for scratch buffer of output values */
364 pScratchOut = pScr2;
365
366 blkCnt = blockSize;
367
368 while(blkCnt > 0u)
369 {
370 /* Perform Multiply-Accumulate */
371 *pScratchOut++ += (q31_t) * px++ * coeff;
372
373 /* Decrement the loop counter */
374 blkCnt--;
375 }
376
377 /* Load the coefficient value and
378 * increment the coefficient buffer for the next set of state values */
379 coeff = *pCoeffs++;
380
381 /* Read Index, from where the state buffer should be read, is calculated. */
382 readIndex = (S->stateIndex - blockSize) - *pTapDelay++;
383
384 /* Wraparound of readIndex */
385 if(readIndex < 0)
386 {
387 readIndex += (int32_t) delaySize;
388 }
389
390 /* Decrement the tap loop counter */
391 tapCnt--;
392 }
393
394 /* All the output values are in pScratchOut buffer.
395 Convert them into 1.15 format, saturate and store in the destination buffer. */
396 /* Loop over the blockSize. */
397 blkCnt = blockSize;
398
399 while(blkCnt > 0u)
400 {
401 *pOut++ = (q15_t) __SSAT(*pScr2++ >> 15, 16);
402 blkCnt--;
403 }
404
405 #endif /* #ifndef ARM_MATH_CM0_FAMILY */
406
407 }
408
409 /**
410 * @} end of FIR_Sparse group
411 */
Imprint / Impressum