/* ----------------------------------------------------------------------
|
* Project: CMSIS DSP Library
|
* Title: arm_fir_fast_q15.c
|
* Description: Q15 Fast FIR filter processing function
|
*
|
* $Date: 27. January 2017
|
* $Revision: V.1.5.1
|
*
|
* Target Processor: Cortex-M cores
|
* -------------------------------------------------------------------- */
|
/*
|
* Copyright (C) 2010-2017 ARM Limited or its affiliates. All rights reserved.
|
*
|
* SPDX-License-Identifier: Apache-2.0
|
*
|
* Licensed under the Apache License, Version 2.0 (the License); you may
|
* not use this file except in compliance with the License.
|
* You may obtain a copy of the License at
|
*
|
* www.apache.org/licenses/LICENSE-2.0
|
*
|
* Unless required by applicable law or agreed to in writing, software
|
* distributed under the License is distributed on an AS IS BASIS, WITHOUT
|
* WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
* See the License for the specific language governing permissions and
|
* limitations under the License.
|
*/
|
|
#include "arm_math.h"
|
|
/**
|
* @ingroup groupFilters
|
*/
|
|
/**
|
* @addtogroup FIR
|
* @{
|
*/
|
|
/**
|
* @param[in] *S points to an instance of the Q15 FIR filter 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 per call.
|
* @return none.
|
*
|
* <b>Scaling and Overflow Behavior:</b>
|
* \par
|
* This fast version uses a 32-bit accumulator with 2.30 format.
|
* The accumulator maintains full precision of the intermediate multiplication results but provides only a single guard bit.
|
* Thus, if the accumulator result overflows it wraps around and distorts the result.
|
* In order to avoid overflows completely the input signal must be scaled down by log2(numTaps) bits.
|
* The 2.30 accumulator is then truncated to 2.15 format and saturated to yield the 1.15 result.
|
*
|
* \par
|
* Refer to the function <code>arm_fir_q15()</code> for a slower implementation of this function which uses 64-bit accumulation to avoid wrap around distortion. Both the slow and the fast versions use the same instance structure.
|
* Use the function <code>arm_fir_init_q15()</code> to initialize the filter structure.
|
*/
|
|
void arm_fir_fast_q15(
|
const arm_fir_instance_q15 * S,
|
q15_t * pSrc,
|
q15_t * pDst,
|
uint32_t blockSize)
|
{
|
q15_t *pState = S->pState; /* State pointer */
|
q15_t *pCoeffs = S->pCoeffs; /* Coefficient pointer */
|
q15_t *pStateCurnt; /* Points to the current sample of the state */
|
q31_t acc0, acc1, acc2, acc3; /* Accumulators */
|
q15_t *pb; /* Temporary pointer for coefficient buffer */
|
q15_t *px; /* Temporary q31 pointer for SIMD state buffer accesses */
|
q31_t x0, x1, x2, c0; /* Temporary variables to hold SIMD state and coefficient values */
|
uint32_t numTaps = S->numTaps; /* Number of taps in the filter */
|
uint32_t tapCnt, blkCnt; /* Loop counters */
|
|
|
/* S->pState points to state array which contains previous frame (numTaps - 1) samples */
|
/* pStateCurnt points to the location where the new input data should be written */
|
pStateCurnt = &(S->pState[(numTaps - 1U)]);
|
|
/* Apply loop unrolling and compute 4 output values simultaneously.
|
* The variables acc0 ... acc3 hold output values that are being computed:
|
*
|
* 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]
|
* 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]
|
* 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]
|
* 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]
|
*/
|
|
blkCnt = blockSize >> 2;
|
|
/* First part of the processing with loop unrolling. Compute 4 outputs at a time.
|
** a second loop below computes the remaining 1 to 3 samples. */
|
while (blkCnt > 0U)
|
{
|
/* Copy four new input samples into the state buffer.
|
** Use 32-bit SIMD to move the 16-bit data. Only requires two copies. */
|
*pStateCurnt++ = *pSrc++;
|
*pStateCurnt++ = *pSrc++;
|
*pStateCurnt++ = *pSrc++;
|
*pStateCurnt++ = *pSrc++;
|
|
|
/* Set all accumulators to zero */
|
acc0 = 0;
|
acc1 = 0;
|
acc2 = 0;
|
acc3 = 0;
|
|
/* Typecast q15_t pointer to q31_t pointer for state reading in q31_t */
|
px = pState;
|
|
/* Typecast q15_t pointer to q31_t pointer for coefficient reading in q31_t */
|
pb = pCoeffs;
|
|
/* Read the first two samples from the state buffer: x[n-N], x[n-N-1] */
|
x0 = *__SIMD32(px)++;
|
|
/* Read the third and forth samples from the state buffer: x[n-N-2], x[n-N-3] */
|
x2 = *__SIMD32(px)++;
|
|
/* Loop over the number of taps. Unroll by a factor of 4.
|
** Repeat until we've computed numTaps-(numTaps%4) coefficients. */
|
tapCnt = numTaps >> 2;
|
|
while (tapCnt > 0)
|
{
|
/* Read the first two coefficients using SIMD: b[N] and b[N-1] coefficients */
|
c0 = *__SIMD32(pb)++;
|
|
/* acc0 += b[N] * x[n-N] + b[N-1] * x[n-N-1] */
|
acc0 = __SMLAD(x0, c0, acc0);
|
|
/* acc2 += b[N] * x[n-N-2] + b[N-1] * x[n-N-3] */
|
acc2 = __SMLAD(x2, c0, acc2);
|
|
/* pack x[n-N-1] and x[n-N-2] */
|
#ifndef ARM_MATH_BIG_ENDIAN
|
x1 = __PKHBT(x2, x0, 0);
|
#else
|
x1 = __PKHBT(x0, x2, 0);
|
#endif
|
|
/* Read state x[n-N-4], x[n-N-5] */
|
x0 = _SIMD32_OFFSET(px);
|
|
/* acc1 += b[N] * x[n-N-1] + b[N-1] * x[n-N-2] */
|
acc1 = __SMLADX(x1, c0, acc1);
|
|
/* pack x[n-N-3] and x[n-N-4] */
|
#ifndef ARM_MATH_BIG_ENDIAN
|
x1 = __PKHBT(x0, x2, 0);
|
#else
|
x1 = __PKHBT(x2, x0, 0);
|
#endif
|
|
/* acc3 += b[N] * x[n-N-3] + b[N-1] * x[n-N-4] */
|
acc3 = __SMLADX(x1, c0, acc3);
|
|
/* Read coefficients b[N-2], b[N-3] */
|
c0 = *__SIMD32(pb)++;
|
|
/* acc0 += b[N-2] * x[n-N-2] + b[N-3] * x[n-N-3] */
|
acc0 = __SMLAD(x2, c0, acc0);
|
|
/* Read state x[n-N-6], x[n-N-7] with offset */
|
x2 = _SIMD32_OFFSET(px + 2U);
|
|
/* acc2 += b[N-2] * x[n-N-4] + b[N-3] * x[n-N-5] */
|
acc2 = __SMLAD(x0, c0, acc2);
|
|
/* acc1 += b[N-2] * x[n-N-3] + b[N-3] * x[n-N-4] */
|
acc1 = __SMLADX(x1, c0, acc1);
|
|
/* pack x[n-N-5] and x[n-N-6] */
|
#ifndef ARM_MATH_BIG_ENDIAN
|
x1 = __PKHBT(x2, x0, 0);
|
#else
|
x1 = __PKHBT(x0, x2, 0);
|
#endif
|
|
/* acc3 += b[N-2] * x[n-N-5] + b[N-3] * x[n-N-6] */
|
acc3 = __SMLADX(x1, c0, acc3);
|
|
/* Update state pointer for next state reading */
|
px += 4U;
|
|
/* Decrement tap count */
|
tapCnt--;
|
|
}
|
|
/* If the filter length is not a multiple of 4, compute the remaining filter taps.
|
** This is always be 2 taps since the filter length is even. */
|
if ((numTaps & 0x3U) != 0U)
|
{
|
|
/* Read last two coefficients */
|
c0 = *__SIMD32(pb)++;
|
|
/* Perform the multiply-accumulates */
|
acc0 = __SMLAD(x0, c0, acc0);
|
acc2 = __SMLAD(x2, c0, acc2);
|
|
/* pack state variables */
|
#ifndef ARM_MATH_BIG_ENDIAN
|
x1 = __PKHBT(x2, x0, 0);
|
#else
|
x1 = __PKHBT(x0, x2, 0);
|
#endif
|
|
/* Read last state variables */
|
x0 = *__SIMD32(px);
|
|
/* Perform the multiply-accumulates */
|
acc1 = __SMLADX(x1, c0, acc1);
|
|
/* pack state variables */
|
#ifndef ARM_MATH_BIG_ENDIAN
|
x1 = __PKHBT(x0, x2, 0);
|
#else
|
x1 = __PKHBT(x2, x0, 0);
|
#endif
|
|
/* Perform the multiply-accumulates */
|
acc3 = __SMLADX(x1, c0, acc3);
|
}
|
|
/* The results in the 4 accumulators are in 2.30 format. Convert to 1.15 with saturation.
|
** Then store the 4 outputs in the destination buffer. */
|
|
#ifndef ARM_MATH_BIG_ENDIAN
|
|
*__SIMD32(pDst)++ =
|
__PKHBT(__SSAT((acc0 >> 15), 16), __SSAT((acc1 >> 15), 16), 16);
|
|
*__SIMD32(pDst)++ =
|
__PKHBT(__SSAT((acc2 >> 15), 16), __SSAT((acc3 >> 15), 16), 16);
|
|
#else
|
|
*__SIMD32(pDst)++ =
|
__PKHBT(__SSAT((acc1 >> 15), 16), __SSAT((acc0 >> 15), 16), 16);
|
|
*__SIMD32(pDst)++ =
|
__PKHBT(__SSAT((acc3 >> 15), 16), __SSAT((acc2 >> 15), 16), 16);
|
|
|
#endif /* #ifndef ARM_MATH_BIG_ENDIAN */
|
|
/* Advance the state pointer by 4 to process the next group of 4 samples */
|
pState = pState + 4U;
|
|
/* Decrement the loop counter */
|
blkCnt--;
|
}
|
|
/* If the blockSize is not a multiple of 4, compute any remaining output samples here.
|
** No loop unrolling is used. */
|
blkCnt = blockSize % 0x4U;
|
while (blkCnt > 0U)
|
{
|
/* Copy two samples into state buffer */
|
*pStateCurnt++ = *pSrc++;
|
|
/* Set the accumulator to zero */
|
acc0 = 0;
|
|
/* Use SIMD to hold states and coefficients */
|
px = pState;
|
pb = pCoeffs;
|
|
tapCnt = numTaps >> 1U;
|
|
do
|
{
|
|
acc0 += (q31_t) * px++ * *pb++;
|
acc0 += (q31_t) * px++ * *pb++;
|
|
tapCnt--;
|
}
|
while (tapCnt > 0U);
|
|
/* The result is in 2.30 format. Convert to 1.15 with saturation.
|
** Then store the output in the destination buffer. */
|
*pDst++ = (q15_t) (__SSAT((acc0 >> 15), 16));
|
|
/* Advance state pointer by 1 for the next sample */
|
pState = pState + 1U;
|
|
/* Decrement the 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;
|
|
/* Calculation of count for copying integer writes */
|
tapCnt = (numTaps - 1U) >> 2;
|
|
while (tapCnt > 0U)
|
{
|
*pStateCurnt++ = *pState++;
|
*pStateCurnt++ = *pState++;
|
*pStateCurnt++ = *pState++;
|
*pStateCurnt++ = *pState++;
|
|
tapCnt--;
|
|
}
|
|
/* Calculation of count for remaining q15_t data */
|
tapCnt = (numTaps - 1U) % 0x4U;
|
|
/* copy remaining data */
|
while (tapCnt > 0U)
|
{
|
*pStateCurnt++ = *pState++;
|
|
/* Decrement the loop counter */
|
tapCnt--;
|
}
|
|
}
|
|
/**
|
* @} end of FIR group
|
*/
|
