diff options
Diffstat (limited to 'Drivers/CMSIS/DSP/Source/FilteringFunctions/arm_fir_decimate_q31.c')
-rw-r--r-- | Drivers/CMSIS/DSP/Source/FilteringFunctions/arm_fir_decimate_q31.c | 354 |
1 files changed, 221 insertions, 133 deletions
diff --git a/Drivers/CMSIS/DSP/Source/FilteringFunctions/arm_fir_decimate_q31.c b/Drivers/CMSIS/DSP/Source/FilteringFunctions/arm_fir_decimate_q31.c index ed6744265..7af8a44e2 100644 --- a/Drivers/CMSIS/DSP/Source/FilteringFunctions/arm_fir_decimate_q31.c +++ b/Drivers/CMSIS/DSP/Source/FilteringFunctions/arm_fir_decimate_q31.c @@ -3,13 +3,13 @@ * Title: arm_fir_decimate_q31.c * Description: Q31 FIR Decimator * - * $Date: 27. January 2017 - * $Revision: V.1.5.1 + * $Date: 18. March 2019 + * $Revision: V1.6.0 * * Target Processor: Cortex-M cores * -------------------------------------------------------------------- */ /* - * Copyright (C) 2010-2017 ARM Limited or its affiliates. All rights reserved. + * Copyright (C) 2010-2019 ARM Limited or its affiliates. All rights reserved. * * SPDX-License-Identifier: Apache-2.0 * @@ -29,130 +29,164 @@ #include "arm_math.h" /** - * @ingroup groupFilters + @ingroup groupFilters */ /** - * @addtogroup FIR_decimate - * @{ + @addtogroup FIR_decimate + @{ */ /** - * @brief Processing function for the Q31 FIR decimator. - * @param[in] *S points to an instance of the Q31 FIR decimator structure. - * @param[in] *pSrc points to the block of input data. - * @param[out] *pDst points to the block of output data - * @param[in] blockSize number of input samples to process per call. - * @return none - * - * <b>Scaling and Overflow Behavior:</b> - * \par - * The function is implemented using an internal 64-bit accumulator. - * The accumulator has a 2.62 format and maintains full precision of the intermediate multiplication results but provides only a single guard bit. - * Thus, if the accumulator result overflows it wraps around rather than clip. - * In order to avoid overflows completely the input signal must be scaled down by log2(numTaps) bits (where log2 is read as log to the base 2). - * After all multiply-accumulates are performed, the 2.62 accumulator is truncated to 1.32 format and then saturated to 1.31 format. - * - * \par - * Refer to the function <code>arm_fir_decimate_fast_q31()</code> for a faster but less precise implementation of this function for Cortex-M3 and Cortex-M4. + @brief Processing function for the Q31 FIR decimator. + @param[in] S points to an instance of the Q31 FIR decimator structure + @param[in] pSrc points to the block of input data + @param[out] pDst points to the block of output data + @param[in] blockSize number of samples to process + @return none + + @par Scaling and Overflow Behavior + The function is implemented using an internal 64-bit accumulator. + The accumulator has a 2.62 format and maintains full precision of the intermediate multiplication results but provides only a single guard bit. + Thus, if the accumulator result overflows it wraps around rather than clip. + In order to avoid overflows completely the input signal must be scaled down by log2(numTaps) bits (where log2 is read as log to the base 2). + After all multiply-accumulates are performed, the 2.62 accumulator is truncated to 1.32 format and then saturated to 1.31 format. + + @remark + Refer to \ref arm_fir_decimate_fast_q31() for a faster but less precise implementation of this function. */ void arm_fir_decimate_q31( const arm_fir_decimate_instance_q31 * S, - q31_t * pSrc, - q31_t * pDst, - uint32_t blockSize) + const q31_t * pSrc, + q31_t * pDst, + uint32_t blockSize) { - q31_t *pState = S->pState; /* State pointer */ - q31_t *pCoeffs = S->pCoeffs; /* Coefficient pointer */ - q31_t *pStateCurnt; /* Points to the current sample of the state */ - q31_t x0, c0; /* Temporary variables to hold state and coefficient values */ - q31_t *px; /* Temporary pointers for state buffer */ - q31_t *pb; /* Temporary pointers for coefficient buffer */ - q63_t sum0; /* Accumulator */ - uint32_t numTaps = S->numTaps; /* Number of taps */ - uint32_t i, tapCnt, blkCnt, outBlockSize = blockSize / S->M; /* Loop counters */ - - -#if defined (ARM_MATH_DSP) - - /* Run the below code for Cortex-M4 and Cortex-M3 */ + q31_t *pState = S->pState; /* State pointer */ + const q31_t *pCoeffs = S->pCoeffs; /* Coefficient pointer */ + q31_t *pStateCur; /* Points to the current sample of the state */ + q31_t *px0; /* Temporary pointer for state buffer */ + const q31_t *pb; /* Temporary pointer for coefficient buffer */ + q31_t x0, c0; /* Temporary variables to hold state and coefficient values */ + q63_t acc0; /* Accumulator */ + uint32_t numTaps = S->numTaps; /* Number of filter coefficients in the filter */ + uint32_t i, tapCnt, blkCnt, outBlockSize = blockSize / S->M; /* Loop counters */ + +#if defined (ARM_MATH_LOOPUNROLL) + q31_t *px1, *px2, *px3; + q31_t x1, x2, x3; + q63_t acc1, acc2, acc3; +#endif /* S->pState buffer contains previous frame (numTaps - 1) samples */ - /* pStateCurnt points to the location where the new input data should be written */ - pStateCurnt = S->pState + (numTaps - 1U); + /* pStateCur points to the location where the new input data should be written */ + pStateCur = S->pState + (numTaps - 1U); - /* Total number of output samples to be computed */ - blkCnt = outBlockSize; +#if defined (ARM_MATH_LOOPUNROLL) + /* Loop unrolling: Compute 4 samples at a time */ + blkCnt = outBlockSize >> 2U; + + /* Samples loop unrolled by 4 */ while (blkCnt > 0U) { - /* Copy decimation factor number of new input samples into the state buffer */ - i = S->M; + /* Copy 4 * decimation factor number of new input samples into the state buffer */ + i = S->M * 4; do { - *pStateCurnt++ = *pSrc++; + *pStateCur++ = *pSrc++; } while (--i); - /* Set accumulator to zero */ - sum0 = 0; + /* Set accumulators to zero */ + acc0 = 0; + acc1 = 0; + acc2 = 0; + acc3 = 0; - /* Initialize state pointer */ - px = pState; + /* Initialize state pointer for all the samples */ + px0 = pState; + px1 = pState + S->M; + px2 = pState + 2 * S->M; + px3 = pState + 3 * S->M; /* Initialize coeff pointer */ pb = pCoeffs; - /* Loop unrolling. Process 4 taps at a time. */ - tapCnt = numTaps >> 2; + /* Loop unrolling: Compute 4 taps at a time */ + tapCnt = numTaps >> 2U; - /* Loop over the number of taps. Unroll by a factor of 4. - ** Repeat until we've computed numTaps-4 coefficients. */ while (tapCnt > 0U) { /* Read the b[numTaps-1] coefficient */ c0 = *(pb++); - /* Read x[n-numTaps-1] sample */ - x0 = *(px++); + /* Read x[n-numTaps-1] sample for acc0 */ + x0 = *(px0++); + /* Read x[n-numTaps-1] sample for acc1 */ + x1 = *(px1++); + /* Read x[n-numTaps-1] sample for acc2 */ + x2 = *(px2++); + /* Read x[n-numTaps-1] sample for acc3 */ + x3 = *(px3++); /* Perform the multiply-accumulate */ - sum0 += (q63_t) x0 *c0; + acc0 += (q63_t) x0 * c0; + acc1 += (q63_t) x1 * c0; + acc2 += (q63_t) x2 * c0; + acc3 += (q63_t) x3 * c0; /* Read the b[numTaps-2] coefficient */ c0 = *(pb++); - /* Read x[n-numTaps-2] sample */ - x0 = *(px++); + /* Read x[n-numTaps-2] sample for acc0, acc1, acc2, acc3 */ + x0 = *(px0++); + x1 = *(px1++); + x2 = *(px2++); + x3 = *(px3++); /* Perform the multiply-accumulate */ - sum0 += (q63_t) x0 *c0; + acc0 += (q63_t) x0 * c0; + acc1 += (q63_t) x1 * c0; + acc2 += (q63_t) x2 * c0; + acc3 += (q63_t) x3 * c0; /* Read the b[numTaps-3] coefficient */ c0 = *(pb++); - /* Read x[n-numTaps-3] sample */ - x0 = *(px++); + /* Read x[n-numTaps-3] sample acc0, acc1, acc2, acc3 */ + x0 = *(px0++); + x1 = *(px1++); + x2 = *(px2++); + x3 = *(px3++); /* Perform the multiply-accumulate */ - sum0 += (q63_t) x0 *c0; + acc0 += (q63_t) x0 * c0; + acc1 += (q63_t) x1 * c0; + acc2 += (q63_t) x2 * c0; + acc3 += (q63_t) x3 * c0; /* Read the b[numTaps-4] coefficient */ c0 = *(pb++); - /* Read x[n-numTaps-4] sample */ - x0 = *(px++); + /* Read x[n-numTaps-4] sample acc0, acc1, acc2, acc3 */ + x0 = *(px0++); + x1 = *(px1++); + x2 = *(px2++); + x3 = *(px3++); /* Perform the multiply-accumulate */ - sum0 += (q63_t) x0 *c0; + acc0 += (q63_t) x0 * c0; + acc1 += (q63_t) x1 * c0; + acc2 += (q63_t) x2 * c0; + acc3 += (q63_t) x3 * c0; - /* Decrement the loop counter */ + /* Decrement loop counter */ tapCnt--; } - /* If the filter length is not a multiple of 4, compute the remaining filter taps */ + /* Loop unrolling: Compute remaining taps */ tapCnt = numTaps % 0x4U; while (tapCnt > 0U) @@ -160,70 +194,46 @@ void arm_fir_decimate_q31( /* Read coefficients */ c0 = *(pb++); - /* Fetch 1 state variable */ - x0 = *(px++); + /* Fetch state variables for acc0, acc1, acc2, acc3 */ + x0 = *(px0++); + x1 = *(px1++); + x2 = *(px2++); + x3 = *(px3++); /* Perform the multiply-accumulate */ - sum0 += (q63_t) x0 *c0; + acc0 += (q63_t) x0 * c0; + acc1 += (q63_t) x1 * c0; + acc2 += (q63_t) x2 * c0; + acc3 += (q63_t) x3 * c0; - /* Decrement the loop counter */ + /* Decrement loop counter */ tapCnt--; } /* Advance the state pointer by the decimation factor * to process the next group of decimation factor number samples */ - pState = pState + S->M; + pState = pState + S->M * 4; /* The result is in the accumulator, store in the destination buffer. */ - *pDst++ = (q31_t) (sum0 >> 31); + *pDst++ = (q31_t) (acc0 >> 31); + *pDst++ = (q31_t) (acc1 >> 31); + *pDst++ = (q31_t) (acc2 >> 31); + *pDst++ = (q31_t) (acc3 >> 31); - /* Decrement the loop counter */ + /* Decrement loop counter */ blkCnt--; } - /* Processing is complete. - ** Now copy the last numTaps - 1 samples to the satrt of the state buffer. - ** This prepares the state buffer for the next function call. */ - - /* Points to the start of the state buffer */ - pStateCurnt = S->pState; - - i = (numTaps - 1U) >> 2U; - - /* copy data */ - while (i > 0U) - { - *pStateCurnt++ = *pState++; - *pStateCurnt++ = *pState++; - *pStateCurnt++ = *pState++; - *pStateCurnt++ = *pState++; - - /* Decrement the loop counter */ - i--; - } - - i = (numTaps - 1U) % 0x04U; - - /* copy data */ - while (i > 0U) - { - *pStateCurnt++ = *pState++; - - /* Decrement the loop counter */ - i--; - } + /* Loop unrolling: Compute remaining samples */ + blkCnt = outBlockSize % 0x4U; #else -/* Run the below code for Cortex-M0 */ - - /* S->pState buffer contains previous frame (numTaps - 1) samples */ - /* pStateCurnt points to the location where the new input data should be written */ - pStateCurnt = S->pState + (numTaps - 1U); - - /* Total number of output samples to be computed */ + /* Initialize blkCnt with number of samples */ blkCnt = outBlockSize; +#endif /* #if defined (ARM_MATH_LOOPUNROLL) */ + while (blkCnt > 0U) { /* Copy decimation factor number of new input samples into the state buffer */ @@ -231,33 +241,88 @@ void arm_fir_decimate_q31( do { - *pStateCurnt++ = *pSrc++; + *pStateCur++ = *pSrc++; } while (--i); /* Set accumulator to zero */ - sum0 = 0; + acc0 = 0; /* Initialize state pointer */ - px = pState; + px0 = pState; /* Initialize coeff pointer */ pb = pCoeffs; +#if defined (ARM_MATH_LOOPUNROLL) + + /* Loop unrolling: Compute 4 taps at a time */ + tapCnt = numTaps >> 2U; + + while (tapCnt > 0U) + { + /* Read the b[numTaps-1] coefficient */ + c0 = *pb++; + + /* Read x[n-numTaps-1] sample */ + x0 = *px0++; + + /* Perform the multiply-accumulate */ + acc0 += (q63_t) x0 * c0; + + /* Read the b[numTaps-2] coefficient */ + c0 = *pb++; + + /* Read x[n-numTaps-2] sample */ + x0 = *px0++; + + /* Perform the multiply-accumulate */ + acc0 += (q63_t) x0 * c0; + + /* Read the b[numTaps-3] coefficient */ + c0 = *pb++; + + /* Read x[n-numTaps-3] sample */ + x0 = *px0++; + + /* Perform the multiply-accumulate */ + acc0 += (q63_t) x0 * c0; + + /* Read the b[numTaps-4] coefficient */ + c0 = *pb++; + + /* Read x[n-numTaps-4] sample */ + x0 = *px0++; + + /* Perform the multiply-accumulate */ + acc0 += (q63_t) x0 * c0; + + /* Decrement loop counter */ + tapCnt--; + } + + /* Loop unrolling: Compute remaining taps */ + tapCnt = numTaps % 0x4U; + +#else + + /* Initialize tapCnt with number of taps */ tapCnt = numTaps; +#endif /* #if defined (ARM_MATH_LOOPUNROLL) */ + while (tapCnt > 0U) { /* Read coefficients */ c0 = *pb++; /* Fetch 1 state variable */ - x0 = *px++; + x0 = *px0++; /* Perform the multiply-accumulate */ - sum0 += (q63_t) x0 *c0; + acc0 += (q63_t) x0 * c0; - /* Decrement the loop counter */ + /* Decrement loop counter */ tapCnt--; } @@ -266,34 +331,57 @@ void arm_fir_decimate_q31( pState = pState + S->M; /* The result is in the accumulator, store in the destination buffer. */ - *pDst++ = (q31_t) (sum0 >> 31); + *pDst++ = (q31_t) (acc0 >> 31); - /* Decrement the loop counter */ + /* Decrement loop counter */ blkCnt--; } /* Processing is complete. - ** Now copy the last numTaps - 1 samples to the start of the state buffer. - ** This prepares the state buffer for the next function call. */ + Now copy the last numTaps - 1 samples to the satrt of the state buffer. + This prepares the state buffer for the next function call. */ /* Points to the start of the state buffer */ - pStateCurnt = S->pState; + pStateCur = S->pState; + +#if defined (ARM_MATH_LOOPUNROLL) - i = numTaps - 1U; + /* Loop unrolling: Compute 4 taps at a time */ + tapCnt = (numTaps - 1U) >> 2U; - /* copy data */ - while (i > 0U) + /* Copy data */ + while (tapCnt > 0U) { - *pStateCurnt++ = *pState++; + *pStateCur++ = *pState++; + *pStateCur++ = *pState++; + *pStateCur++ = *pState++; + *pStateCur++ = *pState++; - /* Decrement the loop counter */ - i--; + /* Decrement loop counter */ + tapCnt--; } -#endif /* #if defined (ARM_MATH_DSP) */ + /* Loop unrolling: Compute remaining taps */ + tapCnt = (numTaps - 1U) % 0x04U; + +#else + + /* Initialize tapCnt with number of taps */ + tapCnt = (numTaps - 1U); + +#endif /* #if defined (ARM_MATH_LOOPUNROLL) */ + + /* Copy data */ + while (tapCnt > 0U) + { + *pStateCur++ = *pState++; + + /* Decrement loop counter */ + tapCnt--; + } } /** - * @} end of FIR_decimate group + @} end of FIR_decimate group */ |