arm_fir_sparse_f32.c

Go to the documentation of this file.

/* ----------------------------------------------------------------------      
* Copyright (C) 2011 ARM Limited. All rights reserved. 
*      
* $Date:        15. December 2011   
* $Revision:    V2.0.0  
*      
* Project:      Cortex-R DSP Library 
* Title:        arm_fir_sparse_f32.c      
*      
* Description:  Floating-point sparse FIR filter processing function.     
*      
* Target Processor:          Cortex-R4/R5
*
* Version 1.0.0 2011/03/08
*     Alpha release.
*
* Version 1.0.1 2011/09/30
*     Beta release.
*
* Version 2.0.0 2011/12/15
*     Final release. 
* 
* ------------------------------------------------------------------- */     
#include "arm_math.h"    
     
void arm_fir_sparse_f32(     
  arm_fir_sparse_instance_f32 * S,     
  float32_t * pSrc,     
  float32_t * pDst,     
  float32_t * pScratchIn,     
  uint32_t blockSize)     
{     
     
  float32_t *pState = S->pState;                 /* State pointer */     
  float32_t *pCoeffs = S->pCoeffs;               /* Coefficient pointer */     
  float32_t *px;                                 /* Scratch buffer pointer */     
  float32_t *py = pState;                        /* Temporary pointers for state buffer */     
  float32_t *pb = pScratchIn;                    /* Temporary pointers for scratch buffer */     
  float32_t *pOut;                               /* Destination pointer */     
  int32_t *pTapDelay = S->pTapDelay;             /* Pointer to the array containing offset of the non-zero tap values. */     
  uint32_t delaySize = S->maxDelay + blockSize;  /* state length */     
  uint16_t numTaps = S->numTaps;                 /* Number of filter coefficients in the filter  */     
  int32_t readIndex;                             /* Read index of the state buffer */     
  uint32_t tapCnt, blkCnt;                       /* loop counters */     
  float32_t coeff = *pCoeffs++;                  /* Read the first coefficient value */     
  float32_t x0, x1;  
  float32_t y0, y1;   
     
     
  /* BlockSize of Input samples are copied into the state buffer */     
  /* StateIndex points to the starting position to write in the state buffer */     
  arm_circularWrite_f32((int32_t *) py, delaySize, &S->stateIndex, 1,     
                        (int32_t *) pSrc, 1, blockSize);     
     
     
  /* Read Index, from where the state buffer should be read, is calculated. */     
  readIndex = ((int32_t) S->stateIndex - (int32_t) blockSize) - *pTapDelay++;     
     
  /* Wraparound of readIndex */     
  if(readIndex < 0)     
  {     
    readIndex += (int32_t) delaySize;     
  }     
     
  /* Working pointer for state buffer is updated */     
  py = pState;     
     
  /* blockSize samples are read from the state buffer */     
  arm_circularRead_f32((int32_t *) py, delaySize, &readIndex, 1,     
                       (int32_t *) pb, (int32_t *) pb, blockSize, 1,     
                       blockSize);     
     
  /* Working pointer for the scratch buffer */     
  px = pb;     
     
  /* Working pointer for destination buffer */     
  pOut = pDst;     
     
  /* Loop over the blockSize. Unroll by a factor of 4.      
   * Compute 4 Multiplications at a time. */     
  blkCnt = blockSize >> 2u;     
     
  while(blkCnt > 0u)     
  {     
    /* Perform Multiplications and store in destination buffer */  
  
    /* Read two values from state */  
    x0 = *px++;  
    x1 = *px++;  
  
    /* multiply  */  
    y0 = x0 * coeff;  
    y1 = x1 * coeff;  
      
    /* Write two values into output buffer */     
    *pOut++ =  y0;     
    *pOut++ =  y1;  
  
    /* Read next two values from state */  
    x0 = *px++;  
    x1 = *px++;  
  
    /* multiply  */  
    y0 = x0 * coeff;  
    y1 = x1 * coeff;  
      
    /* Write next two values into output buffer */     
    *pOut++ =  y0;     
    *pOut++ =  y1;    
     
    /* Decrement the loop counter */     
    blkCnt--;     
  }     
     
  /* If the blockSize is not a multiple of 4,      
   * compute the remaining samples */     
  blkCnt = blockSize % 0x4u;     
     
  while(blkCnt > 0u)     
  {     
    /* Perform Multiplications and store in destination buffer */     
    *pOut++ = *px++ * coeff;     
     
    /* Decrement the loop counter */     
    blkCnt--;     
  }     
     
  /* Load the coefficient value and      
   * increment the coefficient buffer for the next set of state values */     
  coeff = *pCoeffs++;     
     
  /* Read Index, from where the state buffer should be read, is calculated. */     
  readIndex = ((int32_t) S->stateIndex - (int32_t) blockSize) - *pTapDelay++;     
     
  /* Wraparound of readIndex */     
  if(readIndex < 0)     
  {     
    readIndex += (int32_t) delaySize;     
  }     
     
  /* Loop over the number of taps. */     
  tapCnt = (uint32_t) numTaps - 1u;     
     
  while(tapCnt > 0u)     
  {     
     
    /* Working pointer for state buffer is updated */     
    py = pState;     
     
    /* blockSize samples are read from the state buffer */     
    arm_circularRead_f32((int32_t *) py, delaySize, &readIndex, 1,     
                         (int32_t *) pb, (int32_t *) pb, blockSize, 1,     
                         blockSize);     
     
    /* Working pointer for the scratch buffer */     
    px = pb;     
     
    /* Working pointer for destination buffer */     
    pOut = pDst;     
     
    /* Loop over the blockSize. Unroll by a factor of 4.      
     * Compute 4 MACS at a time. */     
    blkCnt = blockSize >> 2u;     
     
    while(blkCnt > 0u)     
    {  
        /* Read two values from scratch */  
        x0 = *px;  
        x1 = *(px + 1u);  
  
        /* Read two values from output buffer */  
        y0 = *pOut;  
        y1 = *(pOut + 1u);  
  
        /* Perform Multiply-Accumulate */  
        y0 = y0 + x0 * coeff;   
        y1 = y1 + x1 * coeff;  
  
        /* Read next two values from scratch */  
        x0 = *(px + 2u);  
        x1 = *(px + 3u);  
  
        /* write multiply-Accumlate values into output buffer */          
        *pOut = y0;  
        *(pOut + 1u) = y1;  
  
        /* Read two values from output buffer */  
        y0 = *(pOut + 2u);  
        y1 = *(pOut + 3u);  
  
        /* Perform Multiply-Accumulate */  
        y0 = y0 + x0 * coeff;   
        y1 = y1 + x1 * coeff;  
  
        /* write multiply-Accumlate values into output buffer */  
        *(pOut + 2u) = y0;  
        *(pOut + 3u) = y1;  
  
        /* update pointers */  
        px += 4u;  
        pOut += 4u;                
    
        /* Decrement the loop counter */     
        blkCnt--;  
             
    }     
     
    /* If the blockSize is not a multiple of 4,      
     * compute the remaining samples */     
    blkCnt = blockSize % 0x4u;     
     
    while(blkCnt > 0u)     
    {     
      /* Read input from scratch */  
      x0 = *px;  
  
      /* Read output value from output buffer */  
      y0 = *pOut;  
  
      /* Perform Multiply-Accumulate */     
      y0 = y0 + x0 * coeff;  
  
      /* update pointers */  
      px += 1u;  
      
      /* write multiply-Accumlate value into output buffer */        
      *pOut++ = y0;     
     
      /* Decrement the loop counter */     
      blkCnt--;     
    }     
     
    /* Load the coefficient value and      
     * increment the coefficient buffer for the next set of state values */     
    coeff = *pCoeffs++;     
     
    /* Read Index, from where the state buffer should be read, is calculated. */     
    readIndex = ((int32_t) S->stateIndex -     
                 (int32_t) blockSize) - *pTapDelay++;     
     
    /* Wraparound of readIndex */     
    if(readIndex < 0)     
    {     
      readIndex += (int32_t) delaySize;     
    }     
     
    /* Decrement the tap loop counter */     
    tapCnt--;     
  }     
     
}