arm_fir_sparse_f32.c

00001 /* ----------------------------------------------------------------------    
00002 * Copyright (C) 2010-2014 ARM Limited. All rights reserved.    
00003 *    
00004 * $Date:        19. March 2015
00005 * $Revision:    V.1.4.5
00006 *    
00007 * Project:      CMSIS DSP Library    
00008 * Title:        arm_fir_sparse_f32.c    
00009 *    
00010 * Description:  Floating-point sparse 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 #include "arm_math.h"
00041 
00042 /**    
00043  * @ingroup groupFilters    
00044  */
00045 
00046 /**    
00047  * @defgroup FIR_Sparse Finite Impulse Response (FIR) Sparse Filters    
00048  *    
00049  * This group of functions implements sparse FIR filters.     
00050  * Sparse FIR filters are equivalent to standard FIR filters except that most of the coefficients are equal to zero.   
00051  * Sparse filters are used for simulating reflections in communications and audio applications.   
00052  *   
00053  * There are separate functions for Q7, Q15, Q31, and floating-point data types.    
00054  * The functions operate on blocks  of input and output data and each call to the function processes    
00055  * <code>blockSize</code> samples through the filter.  <code>pSrc</code> and    
00056  * <code>pDst</code> points to input and output arrays respectively containing <code>blockSize</code> values.    
00057  *    
00058  * \par Algorithm:    
00059  * The sparse filter instant structure contains an array of tap indices <code>pTapDelay</code> which specifies the locations of the non-zero coefficients.   
00060  * This is in addition to the coefficient array <code>b</code>.   
00061  * The implementation essentially skips the multiplications by zero and leads to an efficient realization.   
00062  * <pre>   
00063  *     y[n] = b[0] * x[n-pTapDelay[0]] + b[1] * x[n-pTapDelay[1]] + b[2] * x[n-pTapDelay[2]] + ...+ b[numTaps-1] * x[n-pTapDelay[numTaps-1]]    
00064  * </pre>    
00065  * \par    
00066  * \image html FIRSparse.gif "Sparse FIR filter.  b[n] represents the filter coefficients"   
00067  * \par    
00068  * <code>pCoeffs</code> points to a coefficient array of size <code>numTaps</code>;    
00069  * <code>pTapDelay</code> points to an array of nonzero indices and is also of size <code>numTaps</code>;   
00070  * <code>pState</code> points to a state array of size <code>maxDelay + blockSize</code>, where   
00071  * <code>maxDelay</code> is the largest offset value that is ever used in the <code>pTapDelay</code> array.   
00072  * Some of the processing functions also require temporary working buffers.   
00073  *   
00074  * \par Instance Structure    
00075  * The coefficients and state variables for a filter are stored together in an instance data structure.    
00076  * A separate instance structure must be defined for each filter.    
00077  * Coefficient and offset arrays may be shared among several instances while state variable arrays cannot be shared.    
00078  * There are separate instance structure declarations for each of the 4 supported data types.    
00079  *    
00080  * \par Initialization Functions    
00081  * There is also an associated initialization function for each data type.    
00082  * The initialization function performs the following operations:    
00083  * - Sets the values of the internal structure fields.    
00084  * - Zeros out the values in the state buffer.    
00085  * To do this manually without calling the init function, assign the follow subfields of the instance structure:
00086  * numTaps, pCoeffs, pTapDelay, maxDelay, stateIndex, pState. Also set all of the values in pState to zero. 
00087  *    
00088  * \par    
00089  * Use of the initialization function is optional.    
00090  * However, if the initialization function is used, then the instance structure cannot be placed into a const data section.    
00091  * To place an instance structure into a const data section, the instance structure must be manually initialized.    
00092  * Set the values in the state buffer to zeros before static initialization.    
00093  * The code below statically initializes each of the 4 different data type filter instance structures    
00094  * <pre>    
00095  *arm_fir_sparse_instance_f32 S = {numTaps, 0, pState, pCoeffs, maxDelay, pTapDelay};    
00096  *arm_fir_sparse_instance_q31 S = {numTaps, 0, pState, pCoeffs, maxDelay, pTapDelay};    
00097  *arm_fir_sparse_instance_q15 S = {numTaps, 0, pState, pCoeffs, maxDelay, pTapDelay};    
00098  *arm_fir_sparse_instance_q7 S =  {numTaps, 0, pState, pCoeffs, maxDelay, pTapDelay};    
00099  * </pre>    
00100  * \par    
00101  *    
00102  * \par Fixed-Point Behavior    
00103  * Care must be taken when using the fixed-point versions of the sparse FIR filter functions.    
00104  * In particular, the overflow and saturation behavior of the accumulator used in each function must be considered.    
00105  * Refer to the function specific documentation below for usage guidelines.    
00106  */
00107 
00108 /**    
00109  * @addtogroup FIR_Sparse    
00110  * @{    
00111  */
00112 
00113 /**   
00114  * @brief Processing function for the floating-point sparse FIR filter.   
00115  * @param[in]  *S          points to an instance of the floating-point sparse FIR structure.   
00116  * @param[in]  *pSrc       points to the block of input data.   
00117  * @param[out] *pDst       points to the block of output data   
00118  * @param[in]  *pScratchIn points to a temporary buffer of size blockSize.   
00119  * @param[in]  blockSize   number of input samples to process per call.   
00120  * @return none.   
00121  */
00122 
00123 void arm_fir_sparse_f32(
00124   arm_fir_sparse_instance_f32 * S,
00125   float32_t * pSrc,
00126   float32_t * pDst,
00127   float32_t * pScratchIn,
00128   uint32_t blockSize)
00129 {
00130 
00131   float32_t *pState = S->pState;                 /* State pointer */
00132   float32_t *pCoeffs = S->pCoeffs;               /* Coefficient pointer */
00133   float32_t *px;                                 /* Scratch buffer pointer */
00134   float32_t *py = pState;                        /* Temporary pointers for state buffer */
00135   float32_t *pb = pScratchIn;                    /* Temporary pointers for scratch buffer */
00136   float32_t *pOut;                               /* Destination pointer */
00137   int32_t *pTapDelay = S->pTapDelay;             /* Pointer to the array containing offset of the non-zero tap values. */
00138   uint32_t delaySize = S->maxDelay + blockSize;  /* state length */
00139   uint16_t numTaps = S->numTaps;                 /* Number of filter coefficients in the filter  */
00140   int32_t readIndex;                             /* Read index of the state buffer */
00141   uint32_t tapCnt, blkCnt;                       /* loop counters */
00142   float32_t coeff = *pCoeffs++;                  /* Read the first coefficient value */
00143 
00144 
00145 
00146   /* BlockSize of Input samples are copied into the state buffer */
00147   /* StateIndex points to the starting position to write in the state buffer */
00148   arm_circularWrite_f32((int32_t *) py, delaySize, &S->stateIndex, 1,
00149                         (int32_t *) pSrc, 1, blockSize);
00150 
00151 
00152   /* Read Index, from where the state buffer should be read, is calculated. */
00153   readIndex = ((int32_t) S->stateIndex - (int32_t) blockSize) - *pTapDelay++;
00154 
00155   /* Wraparound of readIndex */
00156   if(readIndex < 0)
00157   {
00158     readIndex += (int32_t) delaySize;
00159   }
00160 
00161   /* Working pointer for state buffer is updated */
00162   py = pState;
00163 
00164   /* blockSize samples are read from the state buffer */
00165   arm_circularRead_f32((int32_t *) py, delaySize, &readIndex, 1,
00166                        (int32_t *) pb, (int32_t *) pb, blockSize, 1,
00167                        blockSize);
00168 
00169   /* Working pointer for the scratch buffer */
00170   px = pb;
00171 
00172   /* Working pointer for destination buffer */
00173   pOut = pDst;
00174 
00175 
00176 #ifndef ARM_MATH_CM0_FAMILY
00177 
00178   /* Run the below code for Cortex-M4 and Cortex-M3 */
00179 
00180   /* Loop over the blockSize. Unroll by a factor of 4.    
00181    * Compute 4 Multiplications at a time. */
00182   blkCnt = blockSize >> 2u;
00183 
00184   while(blkCnt > 0u)
00185   {
00186     /* Perform Multiplications and store in destination buffer */
00187     *pOut++ = *px++ * coeff;
00188     *pOut++ = *px++ * coeff;
00189     *pOut++ = *px++ * coeff;
00190     *pOut++ = *px++ * coeff;
00191 
00192     /* Decrement the loop counter */
00193     blkCnt--;
00194   }
00195 
00196   /* If the blockSize is not a multiple of 4,    
00197    * compute the remaining samples */
00198   blkCnt = blockSize % 0x4u;
00199 
00200   while(blkCnt > 0u)
00201   {
00202     /* Perform Multiplications and store in destination buffer */
00203     *pOut++ = *px++ * coeff;
00204 
00205     /* Decrement the loop counter */
00206     blkCnt--;
00207   }
00208 
00209   /* Load the coefficient value and    
00210    * increment the coefficient buffer for the next set of state values */
00211   coeff = *pCoeffs++;
00212 
00213   /* Read Index, from where the state buffer should be read, is calculated. */
00214   readIndex = ((int32_t) S->stateIndex - (int32_t) blockSize) - *pTapDelay++;
00215 
00216   /* Wraparound of readIndex */
00217   if(readIndex < 0)
00218   {
00219     readIndex += (int32_t) delaySize;
00220   }
00221 
00222   /* Loop over the number of taps. */
00223   tapCnt = (uint32_t) numTaps - 2u;
00224 
00225   while(tapCnt > 0u)
00226   {
00227 
00228     /* Working pointer for state buffer is updated */
00229     py = pState;
00230 
00231     /* blockSize samples are read from the state buffer */
00232     arm_circularRead_f32((int32_t *) py, delaySize, &readIndex, 1,
00233                          (int32_t *) pb, (int32_t *) pb, blockSize, 1,
00234                          blockSize);
00235 
00236     /* Working pointer for the scratch buffer */
00237     px = pb;
00238 
00239     /* Working pointer for destination buffer */
00240     pOut = pDst;
00241 
00242     /* Loop over the blockSize. Unroll by a factor of 4.    
00243      * Compute 4 MACS at a time. */
00244     blkCnt = blockSize >> 2u;
00245 
00246     while(blkCnt > 0u)
00247     {
00248       /* Perform Multiply-Accumulate */
00249       *pOut++ += *px++ * coeff;
00250       *pOut++ += *px++ * coeff;
00251       *pOut++ += *px++ * coeff;
00252       *pOut++ += *px++ * coeff;
00253 
00254       /* Decrement the loop counter */
00255       blkCnt--;
00256     }
00257 
00258     /* If the blockSize is not a multiple of 4,    
00259      * compute the remaining samples */
00260     blkCnt = blockSize % 0x4u;
00261 
00262     while(blkCnt > 0u)
00263     {
00264       /* Perform Multiply-Accumulate */
00265       *pOut++ += *px++ * coeff;
00266 
00267       /* Decrement the loop counter */
00268       blkCnt--;
00269     }
00270 
00271     /* Load the coefficient value and    
00272      * increment the coefficient buffer for the next set of state values */
00273     coeff = *pCoeffs++;
00274 
00275     /* Read Index, from where the state buffer should be read, is calculated. */
00276     readIndex = ((int32_t) S->stateIndex -
00277                  (int32_t) blockSize) - *pTapDelay++;
00278 
00279     /* Wraparound of readIndex */
00280     if(readIndex < 0)
00281     {
00282       readIndex += (int32_t) delaySize;
00283     }
00284 
00285     /* Decrement the tap loop counter */
00286     tapCnt--;
00287   }
00288     
00289     /* Compute last tap without the final read of pTapDelay */
00290 
00291     /* Working pointer for state buffer is updated */
00292     py = pState;
00293 
00294     /* blockSize samples are read from the state buffer */
00295     arm_circularRead_f32((int32_t *) py, delaySize, &readIndex, 1,
00296                                              (int32_t *) pb, (int32_t *) pb, blockSize, 1,
00297                                              blockSize);
00298 
00299     /* Working pointer for the scratch buffer */
00300     px = pb;
00301 
00302     /* Working pointer for destination buffer */
00303     pOut = pDst;
00304 
00305     /* Loop over the blockSize. Unroll by a factor of 4.    
00306      * Compute 4 MACS at a time. */
00307     blkCnt = blockSize >> 2u;
00308 
00309     while(blkCnt > 0u)
00310     {
00311         /* Perform Multiply-Accumulate */
00312         *pOut++ += *px++ * coeff;
00313         *pOut++ += *px++ * coeff;
00314         *pOut++ += *px++ * coeff;
00315         *pOut++ += *px++ * coeff;
00316 
00317         /* Decrement the loop counter */
00318         blkCnt--;
00319     }
00320 
00321     /* If the blockSize is not a multiple of 4,    
00322      * compute the remaining samples */
00323     blkCnt = blockSize % 0x4u;
00324 
00325     while(blkCnt > 0u)
00326     {
00327         /* Perform Multiply-Accumulate */
00328         *pOut++ += *px++ * coeff;
00329 
00330         /* Decrement the loop counter */
00331         blkCnt--;
00332     }
00333 
00334 #else
00335 
00336 /* Run the below code for Cortex-M0 */
00337 
00338   blkCnt = blockSize;
00339 
00340   while(blkCnt > 0u)
00341   {
00342     /* Perform Multiplications and store in destination buffer */
00343     *pOut++ = *px++ * coeff;
00344 
00345     /* Decrement the loop counter */
00346     blkCnt--;
00347   }
00348 
00349   /* Load the coefficient value and           
00350    * increment the coefficient buffer for the next set of state values */
00351   coeff = *pCoeffs++;
00352 
00353   /* Read Index, from where the state buffer should be read, is calculated. */
00354   readIndex = ((int32_t) S->stateIndex - (int32_t) blockSize) - *pTapDelay++;
00355 
00356   /* Wraparound of readIndex */
00357   if(readIndex < 0)
00358   {
00359     readIndex += (int32_t) delaySize;
00360   }
00361 
00362   /* Loop over the number of taps. */
00363   tapCnt = (uint32_t) numTaps - 2u;
00364 
00365   while(tapCnt > 0u)
00366   {
00367 
00368     /* Working pointer for state buffer is updated */
00369     py = pState;
00370 
00371     /* blockSize samples are read from the state buffer */
00372     arm_circularRead_f32((int32_t *) py, delaySize, &readIndex, 1,
00373                          (int32_t *) pb, (int32_t *) pb, blockSize, 1,
00374                          blockSize);
00375 
00376     /* Working pointer for the scratch buffer */
00377     px = pb;
00378 
00379     /* Working pointer for destination buffer */
00380     pOut = pDst;
00381 
00382     blkCnt = blockSize;
00383 
00384     while(blkCnt > 0u)
00385     {
00386       /* Perform Multiply-Accumulate */
00387       *pOut++ += *px++ * coeff;
00388 
00389       /* Decrement the loop counter */
00390       blkCnt--;
00391     }
00392 
00393     /* Load the coefficient value and           
00394      * increment the coefficient buffer for the next set of state values */
00395     coeff = *pCoeffs++;
00396 
00397     /* Read Index, from where the state buffer should be read, is calculated. */
00398     readIndex =
00399       ((int32_t) S->stateIndex - (int32_t) blockSize) - *pTapDelay++;
00400 
00401     /* Wraparound of readIndex */
00402     if(readIndex < 0)
00403     {
00404       readIndex += (int32_t) delaySize;
00405     }
00406 
00407     /* Decrement the tap loop counter */
00408     tapCnt--;
00409   }
00410     
00411     /* Compute last tap without the final read of pTapDelay */  
00412     
00413     /* Working pointer for state buffer is updated */
00414     py = pState;
00415 
00416     /* blockSize samples are read from the state buffer */
00417     arm_circularRead_f32((int32_t *) py, delaySize, &readIndex, 1,
00418                                              (int32_t *) pb, (int32_t *) pb, blockSize, 1,
00419                                              blockSize);
00420 
00421     /* Working pointer for the scratch buffer */
00422     px = pb;
00423 
00424     /* Working pointer for destination buffer */
00425     pOut = pDst;
00426 
00427     blkCnt = blockSize;
00428 
00429     while(blkCnt > 0u)
00430     {
00431         /* Perform Multiply-Accumulate */
00432         *pOut++ += *px++ * coeff;
00433 
00434         /* Decrement the loop counter */
00435         blkCnt--;
00436     }
00437 
00438 #endif /*   #ifndef ARM_MATH_CM0_FAMILY        */
00439 
00440 }
00441 
00442 /**    
00443  * @} end of FIR_Sparse group    
00444  */