tmk_core/tool/mbed/mbed-sdk/libraries/dsp/cmsis_dsp/FilteringFunctions/arm_fir_decimate_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_decimate_q15.c
   9 *
  10 * Description:  Q15 FIR Decimator.
  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
  41 #include "arm_math.h"
  42
  43 /**
  44  * @ingroup groupFilters
  45  */
  46
  47 /**
  48  * @addtogroup FIR_decimate
  49  * @{
  50  */
  51
  52 /**
  53  * @brief Processing function for the Q15 FIR decimator.
  54  * @param[in] *S points to an instance of the Q15 FIR decimator structure.
  55  * @param[in] *pSrc points to the block of input data.
  56  * @param[out] *pDst points to the location where the output result is written.
  57  * @param[in] blockSize number of input samples to process per call.
  58  * @return none.
  59  *
  60  * <b>Scaling and Overflow Behavior:</b>
  61  * \par
  62  * The function is implemented using a 64-bit internal accumulator.
  63  * Both coefficients and state variables are represented in 1.15 format and multiplications yield a 2.30 result.
  64  * The 2.30 intermediate results are accumulated in a 64-bit accumulator in 34.30 format.
  65  * There is no risk of internal overflow with this approach and the full precision of intermediate multiplications is preserved.
  66  * After all additions have been performed, the accumulator is truncated to 34.15 format by discarding low 15 bits.
  67  * Lastly, the accumulator is saturated to yield a result in 1.15 format.
  68  *
  69  * \par
  70  * Refer to the function <code>arm_fir_decimate_fast_q15()</code> for a faster but less precise implementation of this function for Cortex-M3 and Cortex-M4.
  71  */
  72
  73 #ifndef ARM_MATH_CM0_FAMILY
  74
  75 #ifndef UNALIGNED_SUPPORT_DISABLE
  76
  77 void arm_fir_decimate_q15(
  78   const arm_fir_decimate_instance_q15 * S,
  79   q15_t * pSrc,
  80   q15_t * pDst,
  81   uint32_t blockSize)
  82 {
  83   q15_t *pState = S->pState;                     /* State pointer */
  84   q15_t *pCoeffs = S->pCoeffs;                   /* Coefficient pointer */
  85   q15_t *pStateCurnt;                            /* Points to the current sample of the state */
  86   q15_t *px;                                     /* Temporary pointer for state buffer */
  87   q15_t *pb;                                     /* Temporary pointer coefficient buffer */
  88   q31_t x0, x1, c0, c1;                          /* Temporary variables to hold state and coefficient values */
  89   q63_t sum0;                                    /* Accumulators */
  90   q63_t acc0, acc1;
  91   q15_t *px0, *px1;
  92   uint32_t blkCntN3;
  93   uint32_t numTaps = S->numTaps;                 /* Number of taps */
  94   uint32_t i, blkCnt, tapCnt, outBlockSize = blockSize / S->M;  /* Loop counters */
  95
  96
  97   /* S->pState buffer contains previous frame (numTaps - 1) samples */
  98   /* pStateCurnt points to the location where the new input data should be written */
  99   pStateCurnt = S->pState + (numTaps - 1u);
 100
 101
 102   /* Total number of output samples to be computed */
 103   blkCnt = outBlockSize / 2;
 104   blkCntN3 = outBlockSize - (2 * blkCnt);
 105
 106
 107   while(blkCnt > 0u)
 108   {
 109     /* Copy decimation factor number of new input samples into the state buffer */
 110     i = 2 * S->M;
 111
 112     do
 113     {
 114       *pStateCurnt++ = *pSrc++;
 115
 116     } while(--i);
 117
 118     /* Set accumulator to zero */
 119     acc0 = 0;
 120     acc1 = 0;
 121
 122     /* Initialize state pointer */
 123     px0 = pState;
 124
 125     px1 = pState + S->M;
 126
 127
 128     /* Initialize coeff pointer */
 129     pb = pCoeffs;
 130
 131     /* Loop unrolling.  Process 4 taps at a time. */
 132     tapCnt = numTaps >> 2;
 133
 134     /* Loop over the number of taps.  Unroll by a factor of 4.
 135      ** Repeat until we've computed numTaps-4 coefficients. */
 136     while(tapCnt > 0u)
 137     {
 138       /* Read the Read b[numTaps-1] and b[numTaps-2]  coefficients */
 139       c0 = *__SIMD32(pb)++;
 140
 141       /* Read x[n-numTaps-1] and x[n-numTaps-2]sample */
 142       x0 = *__SIMD32(px0)++;
 143
 144       x1 = *__SIMD32(px1)++;
 145
 146       /* Perform the multiply-accumulate */
 147       acc0 = __SMLALD(x0, c0, acc0);
 148
 149       acc1 = __SMLALD(x1, c0, acc1);
 150
 151       /* Read the b[numTaps-3] and b[numTaps-4] coefficient */
 152       c0 = *__SIMD32(pb)++;
 153
 154       /* Read x[n-numTaps-2] and x[n-numTaps-3] sample */
 155       x0 = *__SIMD32(px0)++;
 156
 157       x1 = *__SIMD32(px1)++;
 158
 159       /* Perform the multiply-accumulate */
 160       acc0 = __SMLALD(x0, c0, acc0);
 161
 162       acc1 = __SMLALD(x1, c0, acc1);
 163
 164       /* Decrement the loop counter */
 165       tapCnt--;
 166     }
 167
 168     /* If the filter length is not a multiple of 4, compute the remaining filter taps */
 169     tapCnt = numTaps % 0x4u;
 170
 171     while(tapCnt > 0u)
 172     {
 173       /* Read coefficients */
 174       c0 = *pb++;
 175
 176       /* Fetch 1 state variable */
 177       x0 = *px0++;
 178
 179       x1 = *px1++;
 180
 181       /* Perform the multiply-accumulate */
 182       acc0 = __SMLALD(x0, c0, acc0);
 183       acc1 = __SMLALD(x1, c0, acc1);
 184
 185       /* Decrement the loop counter */
 186       tapCnt--;
 187     }
 188
 189     /* Advance the state pointer by the decimation factor
 190      * to process the next group of decimation factor number samples */
 191     pState = pState + S->M * 2;
 192
 193     /* Store filter output, smlad returns the values in 2.14 format */
 194     /* so downsacle by 15 to get output in 1.15 */
 195     *pDst++ = (q15_t) (__SSAT((acc0 >> 15), 16));
 196     *pDst++ = (q15_t) (__SSAT((acc1 >> 15), 16));
 197
 198     /* Decrement the loop counter */
 199     blkCnt--;
 200   }
 201
 202
 203
 204   while(blkCntN3 > 0u)
 205   {
 206     /* Copy decimation factor number of new input samples into the state buffer */
 207     i = S->M;
 208
 209     do
 210     {
 211       *pStateCurnt++ = *pSrc++;
 212
 213     } while(--i);
 214
 215     /*Set sum to zero */
 216     sum0 = 0;
 217
 218     /* Initialize state pointer */
 219     px = pState;
 220
 221     /* Initialize coeff pointer */
 222     pb = pCoeffs;
 223
 224     /* Loop unrolling.  Process 4 taps at a time. */
 225     tapCnt = numTaps >> 2;
 226
 227     /* Loop over the number of taps.  Unroll by a factor of 4.
 228      ** Repeat until we've computed numTaps-4 coefficients. */
 229     while(tapCnt > 0u)
 230     {
 231       /* Read the Read b[numTaps-1] and b[numTaps-2]  coefficients */
 232       c0 = *__SIMD32(pb)++;
 233
 234       /* Read x[n-numTaps-1] and x[n-numTaps-2]sample */
 235       x0 = *__SIMD32(px)++;
 236
 237       /* Read the b[numTaps-3] and b[numTaps-4] coefficient */
 238       c1 = *__SIMD32(pb)++;
 239
 240       /* Perform the multiply-accumulate */
 241       sum0 = __SMLALD(x0, c0, sum0);
 242
 243       /* Read x[n-numTaps-2] and x[n-numTaps-3] sample */
 244       x0 = *__SIMD32(px)++;
 245
 246       /* Perform the multiply-accumulate */
 247       sum0 = __SMLALD(x0, c1, sum0);
 248
 249       /* Decrement the loop counter */
 250       tapCnt--;
 251     }
 252
 253     /* If the filter length is not a multiple of 4, compute the remaining filter taps */
 254     tapCnt = numTaps % 0x4u;
 255
 256     while(tapCnt > 0u)
 257     {
 258       /* Read coefficients */
 259       c0 = *pb++;
 260
 261       /* Fetch 1 state variable */
 262       x0 = *px++;
 263
 264       /* Perform the multiply-accumulate */
 265       sum0 = __SMLALD(x0, c0, sum0);
 266
 267       /* Decrement the loop counter */
 268       tapCnt--;
 269     }
 270
 271     /* Advance the state pointer by the decimation factor
 272      * to process the next group of decimation factor number samples */
 273     pState = pState + S->M;
 274
 275     /* Store filter output, smlad returns the values in 2.14 format */
 276     /* so downsacle by 15 to get output in 1.15 */
 277     *pDst++ = (q15_t) (__SSAT((sum0 >> 15), 16));
 278
 279     /* Decrement the loop counter */
 280     blkCntN3--;
 281   }
 282
 283   /* Processing is complete.
 284    ** Now copy the last numTaps - 1 samples to the satrt of the state buffer.
 285    ** This prepares the state buffer for the next function call. */
 286
 287   /* Points to the start of the state buffer */
 288   pStateCurnt = S->pState;
 289
 290   i = (numTaps - 1u) >> 2u;
 291
 292   /* copy data */
 293   while(i > 0u)
 294   {
 295     *__SIMD32(pStateCurnt)++ = *__SIMD32(pState)++;
 296     *__SIMD32(pStateCurnt)++ = *__SIMD32(pState)++;
 297
 298     /* Decrement the loop counter */
 299     i--;
 300   }
 301
 302   i = (numTaps - 1u) % 0x04u;
 303
 304   /* copy data */
 305   while(i > 0u)
 306   {
 307     *pStateCurnt++ = *pState++;
 308
 309     /* Decrement the loop counter */
 310     i--;
 311   }
 312 }
 313
 314 #else
 315
 316
 317 void arm_fir_decimate_q15(
 318   const arm_fir_decimate_instance_q15 * S,
 319   q15_t * pSrc,
 320   q15_t * pDst,
 321   uint32_t blockSize)
 322 {
 323   q15_t *pState = S->pState;                     /* State pointer */
 324   q15_t *pCoeffs = S->pCoeffs;                   /* Coefficient pointer */
 325   q15_t *pStateCurnt;                            /* Points to the current sample of the state */
 326   q15_t *px;                                     /* Temporary pointer for state buffer */
 327   q15_t *pb;                                     /* Temporary pointer coefficient buffer */
 328   q15_t x0, x1, c0;                              /* Temporary variables to hold state and coefficient values */
 329   q63_t sum0;                                    /* Accumulators */
 330   q63_t acc0, acc1;
 331   q15_t *px0, *px1;
 332   uint32_t blkCntN3;
 333   uint32_t numTaps = S->numTaps;                 /* Number of taps */
 334   uint32_t i, blkCnt, tapCnt, outBlockSize = blockSize / S->M;  /* Loop counters */
 335
 336
 337   /* S->pState buffer contains previous frame (numTaps - 1) samples */
 338   /* pStateCurnt points to the location where the new input data should be written */
 339   pStateCurnt = S->pState + (numTaps - 1u);
 340
 341
 342   /* Total number of output samples to be computed */
 343   blkCnt = outBlockSize / 2;
 344   blkCntN3 = outBlockSize - (2 * blkCnt);
 345
 346   while(blkCnt > 0u)
 347   {
 348     /* Copy decimation factor number of new input samples into the state buffer */
 349     i = 2 * S->M;
 350
 351     do
 352     {
 353       *pStateCurnt++ = *pSrc++;
 354
 355     } while(--i);
 356
 357     /* Set accumulator to zero */
 358     acc0 = 0;
 359     acc1 = 0;
 360
 361     /* Initialize state pointer */
 362     px0 = pState;
 363
 364     px1 = pState + S->M;
 365
 366
 367     /* Initialize coeff pointer */
 368     pb = pCoeffs;
 369
 370     /* Loop unrolling.  Process 4 taps at a time. */
 371     tapCnt = numTaps >> 2;
 372
 373     /* Loop over the number of taps.  Unroll by a factor of 4.
 374      ** Repeat until we've computed numTaps-4 coefficients. */
 375     while(tapCnt > 0u)
 376     {
 377       /* Read the Read b[numTaps-1] coefficients */
 378       c0 = *pb++;
 379
 380       /* Read x[n-numTaps-1] for sample 0 and for sample 1 */
 381       x0 = *px0++;
 382       x1 = *px1++;
 383
 384       /* Perform the multiply-accumulate */
 385       acc0 += x0 * c0;
 386       acc1 += x1 * c0;
 387
 388       /* Read the b[numTaps-2] coefficient */
 389       c0 = *pb++;
 390
 391       /* Read x[n-numTaps-2] for sample 0 and sample 1 */
 392       x0 = *px0++;
 393       x1 = *px1++;
 394
 395       /* Perform the multiply-accumulate */
 396       acc0 += x0 * c0;
 397       acc1 += x1 * c0;
 398
 399       /* Read the b[numTaps-3] coefficients */
 400       c0 = *pb++;
 401
 402       /* Read x[n-numTaps-3] for sample 0 and sample 1 */
 403       x0 = *px0++;
 404       x1 = *px1++;
 405
 406       /* Perform the multiply-accumulate */
 407       acc0 += x0 * c0;
 408       acc1 += x1 * c0;
 409
 410       /* Read the b[numTaps-4] coefficient */
 411       c0 = *pb++;
 412
 413       /* Read x[n-numTaps-4] for sample 0 and sample 1 */
 414       x0 = *px0++;
 415       x1 = *px1++;
 416
 417       /* Perform the multiply-accumulate */
 418       acc0 += x0 * c0;
 419       acc1 += x1 * c0;
 420
 421       /* Decrement the loop counter */
 422       tapCnt--;
 423     }
 424
 425     /* If the filter length is not a multiple of 4, compute the remaining filter taps */
 426     tapCnt = numTaps % 0x4u;
 427
 428     while(tapCnt > 0u)
 429     {
 430       /* Read coefficients */
 431       c0 = *pb++;
 432
 433       /* Fetch 1 state variable */
 434       x0 = *px0++;
 435       x1 = *px1++;
 436
 437       /* Perform the multiply-accumulate */
 438       acc0 += x0 * c0;
 439       acc1 += x1 * c0;
 440
 441       /* Decrement the loop counter */
 442       tapCnt--;
 443     }
 444
 445     /* Advance the state pointer by the decimation factor
 446      * to process the next group of decimation factor number samples */
 447     pState = pState + S->M * 2;
 448
 449     /* Store filter output, smlad returns the values in 2.14 format */
 450     /* so downsacle by 15 to get output in 1.15 */
 451
 452     *pDst++ = (q15_t) (__SSAT((acc0 >> 15), 16));
 453     *pDst++ = (q15_t) (__SSAT((acc1 >> 15), 16));
 454
 455     /* Decrement the loop counter */
 456     blkCnt--;
 457   }
 458
 459   while(blkCntN3 > 0u)
 460   {
 461     /* Copy decimation factor number of new input samples into the state buffer */
 462     i = S->M;
 463
 464     do
 465     {
 466       *pStateCurnt++ = *pSrc++;
 467
 468     } while(--i);
 469
 470     /*Set sum to zero */
 471     sum0 = 0;
 472
 473     /* Initialize state pointer */
 474     px = pState;
 475
 476     /* Initialize coeff pointer */
 477     pb = pCoeffs;
 478
 479     /* Loop unrolling.  Process 4 taps at a time. */
 480     tapCnt = numTaps >> 2;
 481
 482     /* Loop over the number of taps.  Unroll by a factor of 4.
 483      ** Repeat until we've computed numTaps-4 coefficients. */
 484     while(tapCnt > 0u)
 485     {
 486       /* Read the Read b[numTaps-1] coefficients */
 487       c0 = *pb++;
 488
 489       /* Read x[n-numTaps-1] and sample */
 490       x0 = *px++;
 491
 492       /* Perform the multiply-accumulate */
 493       sum0 += x0 * c0;
 494
 495       /* Read the b[numTaps-2] coefficient */
 496       c0 = *pb++;
 497
 498       /* Read x[n-numTaps-2] and  sample */
 499       x0 = *px++;
 500
 501       /* Perform the multiply-accumulate */
 502       sum0 += x0 * c0;
 503
 504       /* Read the b[numTaps-3]  coefficients */
 505       c0 = *pb++;
 506
 507       /* Read x[n-numTaps-3] sample */
 508       x0 = *px++;
 509
 510       /* Perform the multiply-accumulate */
 511       sum0 += x0 * c0;
 512
 513       /* Read the b[numTaps-4] coefficient */
 514       c0 = *pb++;
 515
 516       /* Read x[n-numTaps-4] sample */
 517       x0 = *px++;
 518
 519       /* Perform the multiply-accumulate */
 520       sum0 += x0 * c0;
 521
 522       /* Decrement the loop counter */
 523       tapCnt--;
 524     }
 525
 526     /* If the filter length is not a multiple of 4, compute the remaining filter taps */
 527     tapCnt = numTaps % 0x4u;
 528
 529     while(tapCnt > 0u)
 530     {
 531       /* Read coefficients */
 532       c0 = *pb++;
 533
 534       /* Fetch 1 state variable */
 535       x0 = *px++;
 536
 537       /* Perform the multiply-accumulate */
 538       sum0 += x0 * c0;
 539
 540       /* Decrement the loop counter */
 541       tapCnt--;
 542     }
 543
 544     /* Advance the state pointer by the decimation factor
 545      * to process the next group of decimation factor number samples */
 546     pState = pState + S->M;
 547
 548     /* Store filter output, smlad returns the values in 2.14 format */
 549     /* so downsacle by 15 to get output in 1.15 */
 550     *pDst++ = (q15_t) (__SSAT((sum0 >> 15), 16));
 551
 552     /* Decrement the loop counter */
 553     blkCntN3--;
 554   }
 555
 556   /* Processing is complete.
 557    ** Now copy the last numTaps - 1 samples to the satrt of the state buffer.
 558    ** This prepares the state buffer for the next function call. */
 559
 560   /* Points to the start of the state buffer */
 561   pStateCurnt = S->pState;
 562
 563   i = (numTaps - 1u) >> 2u;
 564
 565   /* copy data */
 566   while(i > 0u)
 567   {
 568     *pStateCurnt++ = *pState++;
 569     *pStateCurnt++ = *pState++;
 570     *pStateCurnt++ = *pState++;
 571     *pStateCurnt++ = *pState++;
 572
 573     /* Decrement the loop counter */
 574     i--;
 575   }
 576
 577   i = (numTaps - 1u) % 0x04u;
 578
 579   /* copy data */
 580   while(i > 0u)
 581   {
 582     *pStateCurnt++ = *pState++;
 583
 584     /* Decrement the loop counter */
 585     i--;
 586   }
 587 }
 588
 589
 590 #endif  /*      #ifndef UNALIGNED_SUPPORT_DISABLE       */
 591
 592 #else
 593
 594
 595 void arm_fir_decimate_q15(
 596   const arm_fir_decimate_instance_q15 * S,
 597   q15_t * pSrc,
 598   q15_t * pDst,
 599   uint32_t blockSize)
 600 {
 601   q15_t *pState = S->pState;                     /* State pointer */
 602   q15_t *pCoeffs = S->pCoeffs;                   /* Coefficient pointer */
 603   q15_t *pStateCurnt;                            /* Points to the current sample of the state */
 604   q15_t *px;                                     /* Temporary pointer for state buffer */
 605   q15_t *pb;                                     /* Temporary pointer coefficient buffer */
 606   q31_t x0, c0;                                  /* Temporary variables to hold state and coefficient values */
 607   q63_t sum0;                                    /* Accumulators */
 608   uint32_t numTaps = S->numTaps;                 /* Number of taps */
 609   uint32_t i, blkCnt, tapCnt, outBlockSize = blockSize / S->M;  /* Loop counters */
 610
 611
 612
 613 /* Run the below code for Cortex-M0 */
 614
 615   /* S->pState buffer contains previous frame (numTaps - 1) samples */
 616   /* pStateCurnt points to the location where the new input data should be written */
 617   pStateCurnt = S->pState + (numTaps - 1u);
 618
 619   /* Total number of output samples to be computed */
 620   blkCnt = outBlockSize;
 621
 622   while(blkCnt > 0u)
 623   {
 624     /* Copy decimation factor number of new input samples into the state buffer */
 625     i = S->M;
 626
 627     do
 628     {
 629       *pStateCurnt++ = *pSrc++;
 630
 631     } while(--i);
 632
 633     /*Set sum to zero */
 634     sum0 = 0;
 635
 636     /* Initialize state pointer */
 637     px = pState;
 638
 639     /* Initialize coeff pointer */
 640     pb = pCoeffs;
 641
 642     tapCnt = numTaps;
 643
 644     while(tapCnt > 0u)
 645     {
 646       /* Read coefficients */
 647       c0 = *pb++;
 648
 649       /* Fetch 1 state variable */
 650       x0 = *px++;
 651
 652       /* Perform the multiply-accumulate */
 653       sum0 += (q31_t) x0 *c0;
 654
 655       /* Decrement the loop counter */
 656       tapCnt--;
 657     }
 658
 659     /* Advance the state pointer by the decimation factor
 660      * to process the next group of decimation factor number samples */
 661     pState = pState + S->M;
 662
 663     /*Store filter output , smlad will return the values in 2.14 format */
 664     /* so downsacle by 15 to get output in 1.15 */
 665     *pDst++ = (q15_t) (__SSAT((sum0 >> 15), 16));
 666
 667     /* Decrement the loop counter */
 668     blkCnt--;
 669   }
 670
 671   /* Processing is complete.
 672    ** Now copy the last numTaps - 1 samples to the start of the state buffer.
 673    ** This prepares the state buffer for the next function call. */
 674
 675   /* Points to the start of the state buffer */
 676   pStateCurnt = S->pState;
 677
 678   i = numTaps - 1u;
 679
 680   /* copy data */
 681   while(i > 0u)
 682   {
 683     *pStateCurnt++ = *pState++;
 684
 685     /* Decrement the loop counter */
 686     i--;
 687   }
 688
 689
 690 }
 691 #endif /*   #ifndef ARM_MATH_CM0_FAMILY */
 692
 693
 694 /**
 695  * @} end of FIR_decimate group
 696  */