tmk_core/tool/mbed/mbed-sdk/libraries/dsp/cmsis_dsp/FilteringFunctions/arm_fir_q7.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_q7.c
   9 *
  10 * Description:  Q7 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
  41 #include "arm_math.h"
  42
  43 /**
  44  * @ingroup groupFilters
  45  */
  46
  47 /**
  48  * @addtogroup FIR
  49  * @{
  50  */
  51
  52 /**
  53  * @param[in]   *S points to an instance of the Q7 FIR filter structure.
  54  * @param[in]   *pSrc points to the block of input data.
  55  * @param[out]  *pDst points to the block of output data.
  56  * @param[in]   blockSize number of samples to process per call.
  57  * @return      none.
  58  *
  59  * <b>Scaling and Overflow Behavior:</b>
  60  * \par
  61  * The function is implemented using a 32-bit internal accumulator.
  62  * Both coefficients and state variables are represented in 1.7 format and multiplications yield a 2.14 result.
  63  * The 2.14 intermediate results are accumulated in a 32-bit accumulator in 18.14 format.
  64  * There is no risk of internal overflow with this approach and the full precision of intermediate multiplications is preserved.
  65  * The accumulator is converted to 18.7 format by discarding the low 7 bits.
  66  * Finally, the result is truncated to 1.7 format.
  67  */
  68
  69 void arm_fir_q7(
  70   const arm_fir_instance_q7 * S,
  71   q7_t * pSrc,
  72   q7_t * pDst,
  73   uint32_t blockSize)
  74 {
  75
  76 #ifndef ARM_MATH_CM0_FAMILY
  77
  78   /* Run the below code for Cortex-M4 and Cortex-M3 */
  79
  80   q7_t *pState = S->pState;                      /* State pointer */
  81   q7_t *pCoeffs = S->pCoeffs;                    /* Coefficient pointer */
  82   q7_t *pStateCurnt;                             /* Points to the current sample of the state */
  83   q7_t x0, x1, x2, x3;                           /* Temporary variables to hold state */
  84   q7_t c0;                                       /* Temporary variable to hold coefficient value */
  85   q7_t *px;                                      /* Temporary pointer for state */
  86   q7_t *pb;                                      /* Temporary pointer for coefficient buffer */
  87   q31_t acc0, acc1, acc2, acc3;                  /* Accumulators */
  88   uint32_t numTaps = S->numTaps;                 /* Number of filter coefficients in the filter */
  89   uint32_t i, tapCnt, blkCnt;                    /* Loop counters */
  90
  91   /* S->pState points to state array which contains previous frame (numTaps - 1) samples */
  92   /* pStateCurnt points to the location where the new input data should be written */
  93   pStateCurnt = &(S->pState[(numTaps - 1u)]);
  94
  95   /* Apply loop unrolling and compute 4 output values simultaneously.
  96    * The variables acc0 ... acc3 hold output values that are being computed:
  97    *
  98    *    acc0 =  b[numTaps-1] * x[n-numTaps-1] + b[numTaps-2] * x[n-numTaps-2] + b[numTaps-3] * x[n-numTaps-3] +...+ b[0] * x[0]
  99    *    acc1 =  b[numTaps-1] * x[n-numTaps] +   b[numTaps-2] * x[n-numTaps-1] + b[numTaps-3] * x[n-numTaps-2] +...+ b[0] * x[1]
 100    *    acc2 =  b[numTaps-1] * x[n-numTaps+1] + b[numTaps-2] * x[n-numTaps] +   b[numTaps-3] * x[n-numTaps-1] +...+ b[0] * x[2]
 101    *    acc3 =  b[numTaps-1] * x[n-numTaps+2] + b[numTaps-2] * x[n-numTaps+1] + b[numTaps-3] * x[n-numTaps]   +...+ b[0] * x[3]
 102    */
 103   blkCnt = blockSize >> 2;
 104
 105   /* First part of the processing with loop unrolling.  Compute 4 outputs at a time.
 106    ** a second loop below computes the remaining 1 to 3 samples. */
 107   while(blkCnt > 0u)
 108   {
 109     /* Copy four new input samples into the state buffer */
 110     *pStateCurnt++ = *pSrc++;
 111     *pStateCurnt++ = *pSrc++;
 112     *pStateCurnt++ = *pSrc++;
 113     *pStateCurnt++ = *pSrc++;
 114
 115     /* Set all accumulators to zero */
 116     acc0 = 0;
 117     acc1 = 0;
 118     acc2 = 0;
 119     acc3 = 0;
 120
 121     /* Initialize state pointer */
 122     px = pState;
 123
 124     /* Initialize coefficient pointer */
 125     pb = pCoeffs;
 126
 127     /* Read the first three samples from the state buffer:
 128      *  x[n-numTaps], x[n-numTaps-1], x[n-numTaps-2] */
 129     x0 = *(px++);
 130     x1 = *(px++);
 131     x2 = *(px++);
 132
 133     /* Loop unrolling.  Process 4 taps at a time. */
 134     tapCnt = numTaps >> 2;
 135     i = tapCnt;
 136
 137     while(i > 0u)
 138     {
 139       /* Read the b[numTaps] coefficient */
 140       c0 = *(pb++);
 141
 142       /* Read x[n-numTaps-3] sample */
 143       x3 = *(px++);
 144
 145       /* acc0 +=  b[numTaps] * x[n-numTaps] */
 146       acc0 += ((q15_t) x0 * c0);
 147
 148       /* acc1 +=  b[numTaps] * x[n-numTaps-1] */
 149       acc1 += ((q15_t) x1 * c0);
 150
 151       /* acc2 +=  b[numTaps] * x[n-numTaps-2] */
 152       acc2 += ((q15_t) x2 * c0);
 153
 154       /* acc3 +=  b[numTaps] * x[n-numTaps-3] */
 155       acc3 += ((q15_t) x3 * c0);
 156
 157       /* Read the b[numTaps-1] coefficient */
 158       c0 = *(pb++);
 159
 160       /* Read x[n-numTaps-4] sample */
 161       x0 = *(px++);
 162
 163       /* Perform the multiply-accumulates */
 164       acc0 += ((q15_t) x1 * c0);
 165       acc1 += ((q15_t) x2 * c0);
 166       acc2 += ((q15_t) x3 * c0);
 167       acc3 += ((q15_t) x0 * c0);
 168
 169       /* Read the b[numTaps-2] coefficient */
 170       c0 = *(pb++);
 171
 172       /* Read x[n-numTaps-5] sample */
 173       x1 = *(px++);
 174
 175       /* Perform the multiply-accumulates */
 176       acc0 += ((q15_t) x2 * c0);
 177       acc1 += ((q15_t) x3 * c0);
 178       acc2 += ((q15_t) x0 * c0);
 179       acc3 += ((q15_t) x1 * c0);
 180       /* Read the b[numTaps-3] coefficients */
 181       c0 = *(pb++);
 182
 183       /* Read x[n-numTaps-6] sample */
 184       x2 = *(px++);
 185
 186       /* Perform the multiply-accumulates */
 187       acc0 += ((q15_t) x3 * c0);
 188       acc1 += ((q15_t) x0 * c0);
 189       acc2 += ((q15_t) x1 * c0);
 190       acc3 += ((q15_t) x2 * c0);
 191       i--;
 192     }
 193
 194     /* If the filter length is not a multiple of 4, compute the remaining filter taps */
 195
 196     i = numTaps - (tapCnt * 4u);
 197     while(i > 0u)
 198     {
 199       /* Read coefficients */
 200       c0 = *(pb++);
 201
 202       /* Fetch 1 state variable */
 203       x3 = *(px++);
 204
 205       /* Perform the multiply-accumulates */
 206       acc0 += ((q15_t) x0 * c0);
 207       acc1 += ((q15_t) x1 * c0);
 208       acc2 += ((q15_t) x2 * c0);
 209       acc3 += ((q15_t) x3 * c0);
 210
 211       /* Reuse the present sample states for next sample */
 212       x0 = x1;
 213       x1 = x2;
 214       x2 = x3;
 215
 216       /* Decrement the loop counter */
 217       i--;
 218     }
 219
 220     /* Advance the state pointer by 4 to process the next group of 4 samples */
 221     pState = pState + 4;
 222
 223     /* The results in the 4 accumulators are in 2.62 format.  Convert to 1.31
 224      ** Then store the 4 outputs in the destination buffer. */
 225     acc0 = __SSAT((acc0 >> 7u), 8);
 226     *pDst++ = acc0;
 227     acc1 = __SSAT((acc1 >> 7u), 8);
 228     *pDst++ = acc1;
 229     acc2 = __SSAT((acc2 >> 7u), 8);
 230     *pDst++ = acc2;
 231     acc3 = __SSAT((acc3 >> 7u), 8);
 232     *pDst++ = acc3;
 233
 234     /* Decrement the samples loop counter */
 235     blkCnt--;
 236   }
 237
 238
 239   /* If the blockSize is not a multiple of 4, compute any remaining output samples here.
 240    ** No loop unrolling is used. */
 241   blkCnt = blockSize % 4u;
 242
 243   while(blkCnt > 0u)
 244   {
 245     /* Copy one sample at a time into state buffer */
 246     *pStateCurnt++ = *pSrc++;
 247
 248     /* Set the accumulator to zero */
 249     acc0 = 0;
 250
 251     /* Initialize state pointer */
 252     px = pState;
 253
 254     /* Initialize Coefficient pointer */
 255     pb = (pCoeffs);
 256
 257     i = numTaps;
 258
 259     /* Perform the multiply-accumulates */
 260     do
 261     {
 262       acc0 += (q15_t) * (px++) * (*(pb++));
 263       i--;
 264     } while(i > 0u);
 265
 266     /* The result is in 2.14 format.  Convert to 1.7
 267      ** Then store the output in the destination buffer. */
 268     *pDst++ = __SSAT((acc0 >> 7u), 8);
 269
 270     /* Advance state pointer by 1 for the next sample */
 271     pState = pState + 1;
 272
 273     /* Decrement the samples loop counter */
 274     blkCnt--;
 275   }
 276
 277   /* Processing is complete.
 278    ** Now copy the last numTaps - 1 samples to the satrt of the state buffer.
 279    ** This prepares the state buffer for the next function call. */
 280
 281   /* Points to the start of the state buffer */
 282   pStateCurnt = S->pState;
 283
 284   tapCnt = (numTaps - 1u) >> 2u;
 285
 286   /* copy data */
 287   while(tapCnt > 0u)
 288   {
 289     *pStateCurnt++ = *pState++;
 290     *pStateCurnt++ = *pState++;
 291     *pStateCurnt++ = *pState++;
 292     *pStateCurnt++ = *pState++;
 293
 294     /* Decrement the loop counter */
 295     tapCnt--;
 296   }
 297
 298   /* Calculate remaining number of copies */
 299   tapCnt = (numTaps - 1u) % 0x4u;
 300
 301   /* Copy the remaining q31_t data */
 302   while(tapCnt > 0u)
 303   {
 304     *pStateCurnt++ = *pState++;
 305
 306     /* Decrement the loop counter */
 307     tapCnt--;
 308   }
 309
 310 #else
 311
 312 /* Run the below code for Cortex-M0 */
 313
 314   uint32_t numTaps = S->numTaps;                 /* Number of taps in the filter */
 315   uint32_t i, blkCnt;                            /* Loop counters */
 316   q7_t *pState = S->pState;                      /* State pointer */
 317   q7_t *pCoeffs = S->pCoeffs;                    /* Coefficient pointer */
 318   q7_t *px, *pb;                                 /* Temporary pointers to state and coeff */
 319   q31_t acc = 0;                                 /* Accumlator */
 320   q7_t *pStateCurnt;                             /* Points to the current sample of the state */
 321
 322
 323   /* S->pState points to state array which contains previous frame (numTaps - 1) samples */
 324   /* pStateCurnt points to the location where the new input data should be written */
 325   pStateCurnt = S->pState + (numTaps - 1u);
 326
 327   /* Initialize blkCnt with blockSize */
 328   blkCnt = blockSize;
 329
 330   /* Perform filtering upto BlockSize - BlockSize%4  */
 331   while(blkCnt > 0u)
 332   {
 333     /* Copy one sample at a time into state buffer */
 334     *pStateCurnt++ = *pSrc++;
 335
 336     /* Set accumulator to zero */
 337     acc = 0;
 338
 339     /* Initialize state pointer of type q7 */
 340     px = pState;
 341
 342     /* Initialize coeff pointer of type q7 */
 343     pb = pCoeffs;
 344
 345
 346     i = numTaps;
 347
 348     while(i > 0u)
 349     {
 350       /* acc =  b[numTaps-1] * x[n-numTaps-1] + b[numTaps-2] * x[n-numTaps-2] + b[numTaps-3] * x[n-numTaps-3] +...+ b[0] * x[0] */
 351       acc += (q15_t) * px++ * *pb++;
 352       i--;
 353     }
 354
 355     /* Store the 1.7 format filter output in destination buffer */
 356     *pDst++ = (q7_t) __SSAT((acc >> 7), 8);
 357
 358     /* Advance the state pointer by 1 to process the next sample */
 359     pState = pState + 1;
 360
 361     /* Decrement the loop counter */
 362     blkCnt--;
 363   }
 364
 365   /* Processing is complete.
 366    ** Now copy the last numTaps - 1 samples to the satrt of the state buffer.
 367    ** This prepares the state buffer for the next function call. */
 368
 369
 370   /* Points to the start of the state buffer */
 371   pStateCurnt = S->pState;
 372
 373
 374   /* Copy numTaps number of values */
 375   i = (numTaps - 1u);
 376
 377   /* Copy q7_t data */
 378   while(i > 0u)
 379   {
 380     *pStateCurnt++ = *pState++;
 381     i--;
 382   }
 383
 384 #endif /*   #ifndef ARM_MATH_CM0_FAMILY */
 385
 386 }
 387
 388 /**
 389  * @} end of FIR group
 390  */