k60-keil

K60-Keil版本（下载安装MDK4.23）

        /* Read x[7] */
        x3 = *(q31_t *) px++;

        /* Perform the multiply-accumulates */
        acc0 = __SMLALD(x0, c0, acc0);
        acc1 = __SMLALD(x1, c0, acc1);
        acc2 = __SMLALDX(x1, c0, acc2);
        acc3 = __SMLALDX(x3, c0, acc3);
      }

      if(k == 2u)
      {
        /* Read y[4], y[5] */
        c0 = *(pb);

        /* Read x[7], x[8] */
        x3 = *(q31_t *) px++;

        /* Read x[9] */
        x2 = *(q31_t *) px++;

        /* Perform the multiply-accumulates */
        acc0 = __SMLALD(x0, c0, acc0);
        acc1 = __SMLALD(x1, c0, acc1);
        acc2 = __SMLALD(x3, c0, acc2);
        acc3 = __SMLALD(x2, c0, acc3);
      }

      if(k == 3u)
      {
        /* Read y[4], y[5] */
        c0 = *pb++;

        /* Read x[7], x[8] */
        x3 = *(q31_t *) px++;

        /* Read x[9] */
        x2 = *(q31_t *) px++;

        /* Perform the multiply-accumulates */
        acc0 = __SMLALD(x0, c0, acc0);
        acc1 = __SMLALD(x1, c0, acc1);
        acc2 = __SMLALD(x3, c0, acc2);
        acc3 = __SMLALD(x2, c0, acc3);

        /* Read y[6] */
#ifdef  ARM_MATH_BIG_ENDIAN

        c0 = (*pb);
        c0 = c0 & 0xFFFF0000;

#else

        c0 = (q15_t) (*pb);
        c0 = c0 & 0x0000FFFF;

#endif /*      #ifdef  ARM_MATH_BIG_ENDIAN     */
        /* Read x[10] */
        x3 = *(q31_t *) px++;

        /* Perform the multiply-accumulates */
        acc0 = __SMLALDX(x1, c0, acc0);
        acc1 = __SMLALD(x2, c0, acc1);
        acc2 = __SMLALDX(x2, c0, acc2);
        acc3 = __SMLALDX(x3, c0, acc3);
      }

      /* Store the result in the accumulator in the destination buffer. */
      *pOut = (q15_t) (__SSAT(acc0 >> 15, 16));
      /* Destination pointer is updated according to the address modifier, inc */
      pOut += inc;

      *pOut = (q15_t) (__SSAT(acc1 >> 15, 16));
      pOut += inc;

      *pOut = (q15_t) (__SSAT(acc2 >> 15, 16));
      pOut += inc;

      *pOut = (q15_t) (__SSAT(acc3 >> 15, 16));
      pOut += inc;

      /* Increment the count by 4 as 4 output values are computed */
      count += 4u;

      /* Update the inputA and inputB pointers for next MAC calculation */
      px = pIn1 + count;
      py = pIn2;
      pb = (q31_t *) (py);


      /* Decrement the loop counter */
      blkCnt--;
    }

    /* If the blockSize2 is not a multiple of 4, compute any remaining output samples here.   
     ** No loop unrolling is used. */
    blkCnt = blockSize2 % 0x4u;

    while(blkCnt > 0u)
    {
      /* Accumulator is made zero for every iteration */
      sum = 0;

      /* Apply loop unrolling and compute 4 MACs simultaneously. */
      k = srcBLen >> 2u;

      /* First part of the processing with loop unrolling.  Compute 4 MACs at a time.   
       ** a second loop below computes MACs for the remaining 1 to 3 samples. */
      while(k > 0u)
      {
        /* Perform the multiply-accumulates */
        sum += ((q63_t) * px++ * *py++);
        sum += ((q63_t) * px++ * *py++);
        sum += ((q63_t) * px++ * *py++);
        sum += ((q63_t) * px++ * *py++);

        /* Decrement the loop counter */
        k--;
      }

      /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.   
       ** No loop unrolling is used. */
      k = srcBLen % 0x4u;

      while(k > 0u)
      {
        /* Perform the multiply-accumulates */
        sum += ((q63_t) * px++ * *py++);

        /* Decrement the loop counter */
        k--;
      }

      /* Store the result in the accumulator in the destination buffer. */
      *pOut = (q15_t) (__SSAT(sum >> 15, 16));
      /* Destination pointer is updated according to the address modifier, inc */
      pOut += inc;

      /* Increment count by 1, as one output value is computed */
      count++;

      /* Update the inputA and inputB pointers for next MAC calculation */
      px = pIn1 + count;
      py = pIn2;

      /* Decrement the loop counter */
      blkCnt--;
    }
  }
  else
  {
    /* If the srcBLen is not a multiple of 4,   
     * the blockSize2 loop cannot be unrolled by 4 */
    blkCnt = blockSize2;

    while(blkCnt > 0u)
    {
      /* Accumulator is made zero for every iteration */
      sum = 0;

      /* Loop over srcBLen */
      k = srcBLen;

      while(k > 0u)
      {
        /* Perform the multiply-accumulate */
        sum += ((q63_t) * px++ * *py++);

        /* Decrement the loop counter */
        k--;
      }

      /* Store the result in the accumulator in the destination buffer. */
      *pOut = (q15_t) (__SSAT(sum >> 15, 16));
      /* Destination pointer is updated according to the address modifier, inc */
      pOut += inc;

      /* Increment the MAC count */
      count++;

      /* Update the inputA and inputB pointers for next MAC calculation */
      px = pIn1 + count;
      py = pIn2;

      /* Decrement the loop counter */
      blkCnt--;
    }
  }

  /* --------------------------   
   * Initializations of stage3   
   * -------------------------*/

  /* sum += x[srcALen-srcBLen+1] * y[0] + x[srcALen-srcBLen+2] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]   
   * sum += x[srcALen-srcBLen+2] * y[0] + x[srcALen-srcBLen+3] * y[1] +...+ x[srcALen-1] * y[srcBLen-1]   
   * ....   
   * sum +=  x[srcALen-2] * y[0] + x[srcALen-1] * y[1]   
   * sum +=  x[srcALen-1] * y[0]   
   */

  /* In this stage the MAC operations are decreased by 1 for every iteration.   
     The count variable holds the number of MAC operations performed */
  count = srcBLen - 1u;

  /* Working pointer of inputA */
  pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
  px = pSrc1;

  /* Working pointer of inputB */
  py = pIn2;

  /* -------------------   
   * Stage3 process   
   * ------------------*/

  while(blockSize3 > 0u)
  {
    /* Accumulator is made zero for every iteration */
    sum = 0;

    /* Apply loop unrolling and compute 4 MACs simultaneously. */
    k = count >> 2u;

    /* First part of the processing with loop unrolling.  Compute 4 MACs at a time.   
     ** a second loop below computes MACs for the remaining 1 to 3 samples. */
    while(k > 0u)
    {
      /* Perform the multiply-accumulates */
      /* sum += x[srcALen - srcBLen + 4] * y[3] , sum += x[srcALen - srcBLen + 3] * y[2] */
      sum = __SMLALD(*__SIMD32(px)++, *__SIMD32(py)++, sum);
      /* sum += x[srcALen - srcBLen + 2] * y[1] , sum += x[srcALen - srcBLen + 1] * y[0] */
      sum = __SMLALD(*__SIMD32(px)++, *__SIMD32(py)++, sum);

      /* Decrement the loop counter */
      k--;
    }

    /* If the count is not a multiple of 4, compute any remaining MACs here.   
     ** No loop unrolling is used. */
    k = count % 0x4u;

    while(k > 0u)
    {
      /* Perform the multiply-accumulates */
      sum = __SMLALD(*px++, *py++, sum);

      /* Decrement the loop counter */
      k--;
    }

    /* Store the result in the accumulator in the destination buffer. */
    *pOut = (q15_t) (__SSAT((sum >> 15), 16));
    /* Destination pointer is updated according to the address modifier, inc */
    pOut += inc;

    /* Update the inputA and inputB pointers for next MAC calculation */
    px = ++pSrc1;
    py = pIn2;

    /* Decrement the MAC count */
    count--;

    /* Decrement the loop counter */
    blockSize3--;
  }

#else

/* Run the below code for Cortex-M0 */

  q15_t *pIn1 = pSrcA;                           /* inputA pointer               */
  q15_t *pIn2 = pSrcB + (srcBLen - 1u);          /* inputB pointer               */
  q63_t sum;                                     /* Accumulators                  */
  uint32_t i = 0u, j;                            /* loop counters */
  uint32_t inv = 0u;                             /* Reverse order flag */
  uint32_t tot = 0u;                             /* Length */

  /* The algorithm implementation is based on the lengths of the inputs. */
  /* srcB is always made to slide across srcA. */
  /* So srcBLen is always considered as shorter or equal to srcALen */
  /* But CORR(x, y) is reverse of CORR(y, x) */
  /* So, when srcBLen > srcALen, output pointer is made to point to the end of the output buffer */
  /* and a varaible, inv is set to 1 */
  /* If lengths are not equal then zero pad has to be done to  make the two   
   * inputs of same length. But to improve the performance, we include zeroes   
   * in the output instead of zero padding either of the the inputs*/
  /* If srcALen > srcBLen, (srcALen - srcBLen) zeroes has to included in the   
   * starting of the output buffer */
  /* If srcALen < srcBLen, (srcALen - srcBLen) zeroes has to included in the  
   * ending of the output buffer */
  /* Once the zero padding is done the remaining of the output is calcualted  
   * using convolution but with the shorter signal time shifted. */

  /* Calculate the length of the remaining sequence */
  tot = ((srcALen + srcBLen) - 2u);

  if(srcALen > srcBLen)
  {
    /* Calculating the number of zeros to be padded to the output */
    j = srcALen - srcBLen;

    /* Initialise the pointer after zero padding */
    pDst += j;
  }

  else if(srcALen < srcBLen)
  {
    /* Initialization to inputB pointer */
    pIn1 = pSrcB;

    /* Initialization to the end of inputA pointer */
    pIn2 = pSrcA + (srcALen - 1u);

    /* Initialisation of the pointer after zero padding */
    pDst = pDst + tot;

    /* Swapping the lengths */
    j = srcALen;
    srcALen = srcBLen;
    srcBLen = j;

    /* Setting the reverse flag */
    inv = 1;

  }

  /* Loop to calculate convolution for output length number of times */
  for (i = 0u; i <= tot; i++)
  {
    /* Initialize sum with zero to carry on MAC operations */
    sum = 0;

    /* Loop to perform MAC operations according to convolution equation */
    for (j = 0u; j <= i; j++)
    {
      /* Check the array limitations */
      if((((i - j) < srcBLen) && (j < srcALen)))
      {
        /* z[i] += x[i-j] * y[j] */
        sum += ((q31_t) pIn1[j] * pIn2[-((int32_t) i - j)]);
      }
    }
    /* Store the output in the destination buffer */
    if(inv == 1)
      *pDst-- = (q15_t) __SSAT((sum >> 15u), 16u);
    else
      *pDst++ = (q15_t) __SSAT((sum >> 15u), 16u);
  }

#endif /*   #ifndef ARM_MATH_CM0 */

}

/**   
 * @} end of Corr group   
 */