arm_fir_q7.c

00001 /* ----------------------------------------------------------------------    
00002 * Copyright (C) 2010-2013 ARM Limited. All rights reserved.    
00003 *    
00004 * $Date:        17. January 2013
00005 * $Revision:    V1.4.1
00006 *    
00007 * Project:      CMSIS DSP Library    
00008 * Title:        arm_fir_q7.c    
00009 *    
00010 * Description:  Q7 FIR filter processing function.    
00011 *    
00012 * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0
00013 *  
00014 * Redistribution and use in source and binary forms, with or without 
00015 * modification, are permitted provided that the following conditions
00016 * are met:
00017 *   - Redistributions of source code must retain the above copyright
00018 *     notice, this list of conditions and the following disclaimer.
00019 *   - Redistributions in binary form must reproduce the above copyright
00020 *     notice, this list of conditions and the following disclaimer in
00021 *     the documentation and/or other materials provided with the 
00022 *     distribution.
00023 *   - Neither the name of ARM LIMITED nor the names of its contributors
00024 *     may be used to endorse or promote products derived from this
00025 *     software without specific prior written permission.
00026 *
00027 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
00028 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
00029 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
00030 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 
00031 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
00032 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
00033 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
00034 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
00035 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
00036 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
00037 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
00038 * POSSIBILITY OF SUCH DAMAGE.   
00039 * -------------------------------------------------------------------- */
00040 
00041 #include "arm_math.h"
00042 
00043 /**    
00044  * @ingroup groupFilters    
00045  */
00046 
00047 /**    
00048  * @addtogroup FIR    
00049  * @{    
00050  */
00051 
00052 /**    
00053  * @param[in]   *S points to an instance of the Q7 FIR filter structure.    
00054  * @param[in]   *pSrc points to the block of input data.    
00055  * @param[out]  *pDst points to the block of output data.    
00056  * @param[in]   blockSize number of samples to process per call.    
00057  * @return  none.    
00058  *    
00059  * <b>Scaling and Overflow Behavior:</b>    
00060  * \par    
00061  * The function is implemented using a 32-bit internal accumulator.    
00062  * Both coefficients and state variables are represented in 1.7 format and multiplications yield a 2.14 result.    
00063  * The 2.14 intermediate results are accumulated in a 32-bit accumulator in 18.14 format.    
00064  * There is no risk of internal overflow with this approach and the full precision of intermediate multiplications is preserved.    
00065  * The accumulator is converted to 18.7 format by discarding the low 7 bits.    
00066  * Finally, the result is truncated to 1.7 format.    
00067  */
00068 
00069 void arm_fir_q7(
00070   const arm_fir_instance_q7 * S,
00071   q7_t * pSrc,
00072   q7_t * pDst,
00073   uint32_t blockSize)
00074 {
00075 
00076 #ifndef ARM_MATH_CM0_FAMILY
00077 
00078   /* Run the below code for Cortex-M4 and Cortex-M3 */
00079 
00080   q7_t *pState = S->pState;                      /* State pointer */
00081   q7_t *pCoeffs = S->pCoeffs;                    /* Coefficient pointer */
00082   q7_t *pStateCurnt;                             /* Points to the current sample of the state */
00083   q7_t x0, x1, x2, x3;                           /* Temporary variables to hold state */
00084   q7_t c0;                                       /* Temporary variable to hold coefficient value */
00085   q7_t *px;                                      /* Temporary pointer for state */
00086   q7_t *pb;                                      /* Temporary pointer for coefficient buffer */
00087   q31_t acc0, acc1, acc2, acc3;                  /* Accumulators */
00088   uint32_t numTaps = S->numTaps;                 /* Number of filter coefficients in the filter */
00089   uint32_t i, tapCnt, blkCnt;                    /* Loop counters */
00090 
00091   /* S->pState points to state array which contains previous frame (numTaps - 1) samples */
00092   /* pStateCurnt points to the location where the new input data should be written */
00093   pStateCurnt = &(S->pState[(numTaps - 1u)]);
00094 
00095   /* Apply loop unrolling and compute 4 output values simultaneously.    
00096    * The variables acc0 ... acc3 hold output values that are being computed:    
00097    *    
00098    *    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]    
00099    *    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]    
00100    *    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]    
00101    *    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]    
00102    */
00103   blkCnt = blockSize >> 2;
00104 
00105   /* First part of the processing with loop unrolling.  Compute 4 outputs at a time.    
00106    ** a second loop below computes the remaining 1 to 3 samples. */
00107   while(blkCnt > 0u)
00108   {
00109     /* Copy four new input samples into the state buffer */
00110     *pStateCurnt++ = *pSrc++;
00111     *pStateCurnt++ = *pSrc++;
00112     *pStateCurnt++ = *pSrc++;
00113     *pStateCurnt++ = *pSrc++;
00114 
00115     /* Set all accumulators to zero */
00116     acc0 = 0;
00117     acc1 = 0;
00118     acc2 = 0;
00119     acc3 = 0;
00120 
00121     /* Initialize state pointer */
00122     px = pState;
00123 
00124     /* Initialize coefficient pointer */
00125     pb = pCoeffs;
00126 
00127     /* Read the first three samples from the state buffer:    
00128      *  x[n-numTaps], x[n-numTaps-1], x[n-numTaps-2] */
00129     x0 = *(px++);
00130     x1 = *(px++);
00131     x2 = *(px++);
00132 
00133     /* Loop unrolling.  Process 4 taps at a time. */
00134     tapCnt = numTaps >> 2;
00135     i = tapCnt;
00136 
00137     while(i > 0u)
00138     {
00139       /* Read the b[numTaps] coefficient */
00140       c0 = *(pb++);
00141 
00142       /* Read x[n-numTaps-3] sample */
00143       x3 = *(px++);
00144 
00145       /* acc0 +=  b[numTaps] * x[n-numTaps] */
00146       acc0 += ((q15_t) x0 * c0);
00147 
00148       /* acc1 +=  b[numTaps] * x[n-numTaps-1] */
00149       acc1 += ((q15_t) x1 * c0);
00150 
00151       /* acc2 +=  b[numTaps] * x[n-numTaps-2] */
00152       acc2 += ((q15_t) x2 * c0);
00153 
00154       /* acc3 +=  b[numTaps] * x[n-numTaps-3] */
00155       acc3 += ((q15_t) x3 * c0);
00156 
00157       /* Read the b[numTaps-1] coefficient */
00158       c0 = *(pb++);
00159 
00160       /* Read x[n-numTaps-4] sample */
00161       x0 = *(px++);
00162 
00163       /* Perform the multiply-accumulates */
00164       acc0 += ((q15_t) x1 * c0);
00165       acc1 += ((q15_t) x2 * c0);
00166       acc2 += ((q15_t) x3 * c0);
00167       acc3 += ((q15_t) x0 * c0);
00168 
00169       /* Read the b[numTaps-2] coefficient */
00170       c0 = *(pb++);
00171 
00172       /* Read x[n-numTaps-5] sample */
00173       x1 = *(px++);
00174 
00175       /* Perform the multiply-accumulates */
00176       acc0 += ((q15_t) x2 * c0);
00177       acc1 += ((q15_t) x3 * c0);
00178       acc2 += ((q15_t) x0 * c0);
00179       acc3 += ((q15_t) x1 * c0);
00180       /* Read the b[numTaps-3] coefficients */
00181       c0 = *(pb++);
00182 
00183       /* Read x[n-numTaps-6] sample */
00184       x2 = *(px++);
00185 
00186       /* Perform the multiply-accumulates */
00187       acc0 += ((q15_t) x3 * c0);
00188       acc1 += ((q15_t) x0 * c0);
00189       acc2 += ((q15_t) x1 * c0);
00190       acc3 += ((q15_t) x2 * c0);
00191       i--;
00192     }
00193 
00194     /* If the filter length is not a multiple of 4, compute the remaining filter taps */
00195 
00196     i = numTaps - (tapCnt * 4u);
00197     while(i > 0u)
00198     {
00199       /* Read coefficients */
00200       c0 = *(pb++);
00201 
00202       /* Fetch 1 state variable */
00203       x3 = *(px++);
00204 
00205       /* Perform the multiply-accumulates */
00206       acc0 += ((q15_t) x0 * c0);
00207       acc1 += ((q15_t) x1 * c0);
00208       acc2 += ((q15_t) x2 * c0);
00209       acc3 += ((q15_t) x3 * c0);
00210 
00211       /* Reuse the present sample states for next sample */
00212       x0 = x1;
00213       x1 = x2;
00214       x2 = x3;
00215 
00216       /* Decrement the loop counter */
00217       i--;
00218     }
00219 
00220     /* Advance the state pointer by 4 to process the next group of 4 samples */
00221     pState = pState + 4;
00222 
00223     /* The results in the 4 accumulators are in 2.62 format.  Convert to 1.31    
00224      ** Then store the 4 outputs in the destination buffer. */
00225     acc0 = __SSAT((acc0 >> 7u), 8);
00226     *pDst++ = acc0;
00227     acc1 = __SSAT((acc1 >> 7u), 8);
00228     *pDst++ = acc1;
00229     acc2 = __SSAT((acc2 >> 7u), 8);
00230     *pDst++ = acc2;
00231     acc3 = __SSAT((acc3 >> 7u), 8);
00232     *pDst++ = acc3;
00233 
00234     /* Decrement the samples loop counter */
00235     blkCnt--;
00236   }
00237 
00238 
00239   /* If the blockSize is not a multiple of 4, compute any remaining output samples here.    
00240    ** No loop unrolling is used. */
00241   blkCnt = blockSize % 4u;
00242 
00243   while(blkCnt > 0u)
00244   {
00245     /* Copy one sample at a time into state buffer */
00246     *pStateCurnt++ = *pSrc++;
00247 
00248     /* Set the accumulator to zero */
00249     acc0 = 0;
00250 
00251     /* Initialize state pointer */
00252     px = pState;
00253 
00254     /* Initialize Coefficient pointer */
00255     pb = (pCoeffs);
00256 
00257     i = numTaps;
00258 
00259     /* Perform the multiply-accumulates */
00260     do
00261     {
00262       acc0 += (q15_t) * (px++) * (*(pb++));
00263       i--;
00264     } while(i > 0u);
00265 
00266     /* The result is in 2.14 format.  Convert to 1.7    
00267      ** Then store the output in the destination buffer. */
00268     *pDst++ = __SSAT((acc0 >> 7u), 8);
00269 
00270     /* Advance state pointer by 1 for the next sample */
00271     pState = pState + 1;
00272 
00273     /* Decrement the samples loop counter */
00274     blkCnt--;
00275   }
00276 
00277   /* Processing is complete.    
00278    ** Now copy the last numTaps - 1 samples to the satrt of the state buffer.    
00279    ** This prepares the state buffer for the next function call. */
00280 
00281   /* Points to the start of the state buffer */
00282   pStateCurnt = S->pState;
00283 
00284   tapCnt = (numTaps - 1u) >> 2u;
00285 
00286   /* copy data */
00287   while(tapCnt > 0u)
00288   {
00289     *pStateCurnt++ = *pState++;
00290     *pStateCurnt++ = *pState++;
00291     *pStateCurnt++ = *pState++;
00292     *pStateCurnt++ = *pState++;
00293 
00294     /* Decrement the loop counter */
00295     tapCnt--;
00296   }
00297 
00298   /* Calculate remaining number of copies */
00299   tapCnt = (numTaps - 1u) % 0x4u;
00300 
00301   /* Copy the remaining q31_t data */
00302   while(tapCnt > 0u)
00303   {
00304     *pStateCurnt++ = *pState++;
00305 
00306     /* Decrement the loop counter */
00307     tapCnt--;
00308   }
00309 
00310 #else
00311 
00312 /* Run the below code for Cortex-M0 */
00313 
00314   uint32_t numTaps = S->numTaps;                 /* Number of taps in the filter */
00315   uint32_t i, blkCnt;                            /* Loop counters */
00316   q7_t *pState = S->pState;                      /* State pointer */
00317   q7_t *pCoeffs = S->pCoeffs;                    /* Coefficient pointer */
00318   q7_t *px, *pb;                                 /* Temporary pointers to state and coeff */
00319   q31_t acc = 0;                                 /* Accumlator */
00320   q7_t *pStateCurnt;                             /* Points to the current sample of the state */
00321 
00322 
00323   /* S->pState points to state array which contains previous frame (numTaps - 1) samples */
00324   /* pStateCurnt points to the location where the new input data should be written */
00325   pStateCurnt = S->pState + (numTaps - 1u);
00326 
00327   /* Initialize blkCnt with blockSize */
00328   blkCnt = blockSize;
00329 
00330   /* Perform filtering upto BlockSize - BlockSize%4  */
00331   while(blkCnt > 0u)
00332   {
00333     /* Copy one sample at a time into state buffer */
00334     *pStateCurnt++ = *pSrc++;
00335 
00336     /* Set accumulator to zero */
00337     acc = 0;
00338 
00339     /* Initialize state pointer of type q7 */
00340     px = pState;
00341 
00342     /* Initialize coeff pointer of type q7 */
00343     pb = pCoeffs;
00344 
00345 
00346     i = numTaps;
00347 
00348     while(i > 0u)
00349     {
00350       /* 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] */
00351       acc += (q15_t) * px++ * *pb++;
00352       i--;
00353     }
00354 
00355     /* Store the 1.7 format filter output in destination buffer */
00356     *pDst++ = (q7_t) __SSAT((acc >> 7), 8);
00357 
00358     /* Advance the state pointer by 1 to process the next sample */
00359     pState = pState + 1;
00360 
00361     /* Decrement the loop counter */
00362     blkCnt--;
00363   }
00364 
00365   /* Processing is complete.         
00366    ** Now copy the last numTaps - 1 samples to the satrt of the state buffer.       
00367    ** This prepares the state buffer for the next function call. */
00368 
00369 
00370   /* Points to the start of the state buffer */
00371   pStateCurnt = S->pState;
00372 
00373 
00374   /* Copy numTaps number of values */
00375   i = (numTaps - 1u);
00376 
00377   /* Copy q7_t data */
00378   while(i > 0u)
00379   {
00380     *pStateCurnt++ = *pState++;
00381     i--;
00382   }
00383 
00384 #endif /*   #ifndef ARM_MATH_CM0_FAMILY */
00385 
00386 }
00387 
00388 /**    
00389  * @} end of FIR group    
00390  */