V4.0.1 of the ARM CMSIS DSP libraries. Note that arm_bitreversal2.s, arm_cfft_f32.c and arm_rfft_fast_f32.c had to be removed. arm_bitreversal2.s will not assemble with the online tools. So, the fast f32 FFT functions are not yet available. All the other FFT functions are available.
Dependents: MPU9150_Example fir_f32 fir_f32 MPU9150_nucleo_noni2cdev ... more
FilteringFunctions/arm_fir_q15.c@0:3d9c67d97d6f, 2014-07-28 (annotated)
- Committer:
- emh203
- Date:
- Mon Jul 28 15:03:15 2014 +0000
- Revision:
- 0:3d9c67d97d6f
1st working commit. Had to remove arm_bitreversal2.s arm_cfft_f32.c and arm_rfft_fast_f32.c. The .s will not assemble. For now I removed these functions so we could at least have a library for the other functions.
Who changed what in which revision?
User | Revision | Line number | New contents of line |
---|---|---|---|
emh203 | 0:3d9c67d97d6f | 1 | /* ---------------------------------------------------------------------- |
emh203 | 0:3d9c67d97d6f | 2 | * Copyright (C) 2010-2014 ARM Limited. All rights reserved. |
emh203 | 0:3d9c67d97d6f | 3 | * |
emh203 | 0:3d9c67d97d6f | 4 | * $Date: 12. March 2014 |
emh203 | 0:3d9c67d97d6f | 5 | * $Revision: V1.4.3 |
emh203 | 0:3d9c67d97d6f | 6 | * |
emh203 | 0:3d9c67d97d6f | 7 | * Project: CMSIS DSP Library |
emh203 | 0:3d9c67d97d6f | 8 | * Title: arm_fir_q15.c |
emh203 | 0:3d9c67d97d6f | 9 | * |
emh203 | 0:3d9c67d97d6f | 10 | * Description: Q15 FIR filter processing function. |
emh203 | 0:3d9c67d97d6f | 11 | * |
emh203 | 0:3d9c67d97d6f | 12 | * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0 |
emh203 | 0:3d9c67d97d6f | 13 | * |
emh203 | 0:3d9c67d97d6f | 14 | * Redistribution and use in source and binary forms, with or without |
emh203 | 0:3d9c67d97d6f | 15 | * modification, are permitted provided that the following conditions |
emh203 | 0:3d9c67d97d6f | 16 | * are met: |
emh203 | 0:3d9c67d97d6f | 17 | * - Redistributions of source code must retain the above copyright |
emh203 | 0:3d9c67d97d6f | 18 | * notice, this list of conditions and the following disclaimer. |
emh203 | 0:3d9c67d97d6f | 19 | * - Redistributions in binary form must reproduce the above copyright |
emh203 | 0:3d9c67d97d6f | 20 | * notice, this list of conditions and the following disclaimer in |
emh203 | 0:3d9c67d97d6f | 21 | * the documentation and/or other materials provided with the |
emh203 | 0:3d9c67d97d6f | 22 | * distribution. |
emh203 | 0:3d9c67d97d6f | 23 | * - Neither the name of ARM LIMITED nor the names of its contributors |
emh203 | 0:3d9c67d97d6f | 24 | * may be used to endorse or promote products derived from this |
emh203 | 0:3d9c67d97d6f | 25 | * software without specific prior written permission. |
emh203 | 0:3d9c67d97d6f | 26 | * |
emh203 | 0:3d9c67d97d6f | 27 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
emh203 | 0:3d9c67d97d6f | 28 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
emh203 | 0:3d9c67d97d6f | 29 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS |
emh203 | 0:3d9c67d97d6f | 30 | * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE |
emh203 | 0:3d9c67d97d6f | 31 | * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, |
emh203 | 0:3d9c67d97d6f | 32 | * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, |
emh203 | 0:3d9c67d97d6f | 33 | * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
emh203 | 0:3d9c67d97d6f | 34 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER |
emh203 | 0:3d9c67d97d6f | 35 | * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT |
emh203 | 0:3d9c67d97d6f | 36 | * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN |
emh203 | 0:3d9c67d97d6f | 37 | * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
emh203 | 0:3d9c67d97d6f | 38 | * POSSIBILITY OF SUCH DAMAGE. |
emh203 | 0:3d9c67d97d6f | 39 | * -------------------------------------------------------------------- */ |
emh203 | 0:3d9c67d97d6f | 40 | |
emh203 | 0:3d9c67d97d6f | 41 | #include "arm_math.h" |
emh203 | 0:3d9c67d97d6f | 42 | |
emh203 | 0:3d9c67d97d6f | 43 | /** |
emh203 | 0:3d9c67d97d6f | 44 | * @ingroup groupFilters |
emh203 | 0:3d9c67d97d6f | 45 | */ |
emh203 | 0:3d9c67d97d6f | 46 | |
emh203 | 0:3d9c67d97d6f | 47 | /** |
emh203 | 0:3d9c67d97d6f | 48 | * @addtogroup FIR |
emh203 | 0:3d9c67d97d6f | 49 | * @{ |
emh203 | 0:3d9c67d97d6f | 50 | */ |
emh203 | 0:3d9c67d97d6f | 51 | |
emh203 | 0:3d9c67d97d6f | 52 | /** |
emh203 | 0:3d9c67d97d6f | 53 | * @brief Processing function for the Q15 FIR filter. |
emh203 | 0:3d9c67d97d6f | 54 | * @param[in] *S points to an instance of the Q15 FIR structure. |
emh203 | 0:3d9c67d97d6f | 55 | * @param[in] *pSrc points to the block of input data. |
emh203 | 0:3d9c67d97d6f | 56 | * @param[out] *pDst points to the block of output data. |
emh203 | 0:3d9c67d97d6f | 57 | * @param[in] blockSize number of samples to process per call. |
emh203 | 0:3d9c67d97d6f | 58 | * @return none. |
emh203 | 0:3d9c67d97d6f | 59 | * |
emh203 | 0:3d9c67d97d6f | 60 | * |
emh203 | 0:3d9c67d97d6f | 61 | * \par Restrictions |
emh203 | 0:3d9c67d97d6f | 62 | * If the silicon does not support unaligned memory access enable the macro UNALIGNED_SUPPORT_DISABLE |
emh203 | 0:3d9c67d97d6f | 63 | * In this case input, output, state buffers should be aligned by 32-bit |
emh203 | 0:3d9c67d97d6f | 64 | * |
emh203 | 0:3d9c67d97d6f | 65 | * <b>Scaling and Overflow Behavior:</b> |
emh203 | 0:3d9c67d97d6f | 66 | * \par |
emh203 | 0:3d9c67d97d6f | 67 | * The function is implemented using a 64-bit internal accumulator. |
emh203 | 0:3d9c67d97d6f | 68 | * Both coefficients and state variables are represented in 1.15 format and multiplications yield a 2.30 result. |
emh203 | 0:3d9c67d97d6f | 69 | * The 2.30 intermediate results are accumulated in a 64-bit accumulator in 34.30 format. |
emh203 | 0:3d9c67d97d6f | 70 | * There is no risk of internal overflow with this approach and the full precision of intermediate multiplications is preserved. |
emh203 | 0:3d9c67d97d6f | 71 | * After all additions have been performed, the accumulator is truncated to 34.15 format by discarding low 15 bits. |
emh203 | 0:3d9c67d97d6f | 72 | * Lastly, the accumulator is saturated to yield a result in 1.15 format. |
emh203 | 0:3d9c67d97d6f | 73 | * |
emh203 | 0:3d9c67d97d6f | 74 | * \par |
emh203 | 0:3d9c67d97d6f | 75 | * Refer to the function <code>arm_fir_fast_q15()</code> for a faster but less precise implementation of this function. |
emh203 | 0:3d9c67d97d6f | 76 | */ |
emh203 | 0:3d9c67d97d6f | 77 | |
emh203 | 0:3d9c67d97d6f | 78 | #ifndef ARM_MATH_CM0_FAMILY |
emh203 | 0:3d9c67d97d6f | 79 | |
emh203 | 0:3d9c67d97d6f | 80 | /* Run the below code for Cortex-M4 and Cortex-M3 */ |
emh203 | 0:3d9c67d97d6f | 81 | |
emh203 | 0:3d9c67d97d6f | 82 | #ifndef UNALIGNED_SUPPORT_DISABLE |
emh203 | 0:3d9c67d97d6f | 83 | |
emh203 | 0:3d9c67d97d6f | 84 | |
emh203 | 0:3d9c67d97d6f | 85 | void arm_fir_q15( |
emh203 | 0:3d9c67d97d6f | 86 | const arm_fir_instance_q15 * S, |
emh203 | 0:3d9c67d97d6f | 87 | q15_t * pSrc, |
emh203 | 0:3d9c67d97d6f | 88 | q15_t * pDst, |
emh203 | 0:3d9c67d97d6f | 89 | uint32_t blockSize) |
emh203 | 0:3d9c67d97d6f | 90 | { |
emh203 | 0:3d9c67d97d6f | 91 | q15_t *pState = S->pState; /* State pointer */ |
emh203 | 0:3d9c67d97d6f | 92 | q15_t *pCoeffs = S->pCoeffs; /* Coefficient pointer */ |
emh203 | 0:3d9c67d97d6f | 93 | q15_t *pStateCurnt; /* Points to the current sample of the state */ |
emh203 | 0:3d9c67d97d6f | 94 | q15_t *px1; /* Temporary q15 pointer for state buffer */ |
emh203 | 0:3d9c67d97d6f | 95 | q15_t *pb; /* Temporary pointer for coefficient buffer */ |
emh203 | 0:3d9c67d97d6f | 96 | q31_t x0, x1, x2, x3, c0; /* Temporary variables to hold SIMD state and coefficient values */ |
emh203 | 0:3d9c67d97d6f | 97 | q63_t acc0, acc1, acc2, acc3; /* Accumulators */ |
emh203 | 0:3d9c67d97d6f | 98 | uint32_t numTaps = S->numTaps; /* Number of taps in the filter */ |
emh203 | 0:3d9c67d97d6f | 99 | uint32_t tapCnt, blkCnt; /* Loop counters */ |
emh203 | 0:3d9c67d97d6f | 100 | |
emh203 | 0:3d9c67d97d6f | 101 | |
emh203 | 0:3d9c67d97d6f | 102 | /* S->pState points to state array which contains previous frame (numTaps - 1) samples */ |
emh203 | 0:3d9c67d97d6f | 103 | /* pStateCurnt points to the location where the new input data should be written */ |
emh203 | 0:3d9c67d97d6f | 104 | pStateCurnt = &(S->pState[(numTaps - 1u)]); |
emh203 | 0:3d9c67d97d6f | 105 | |
emh203 | 0:3d9c67d97d6f | 106 | /* Apply loop unrolling and compute 4 output values simultaneously. |
emh203 | 0:3d9c67d97d6f | 107 | * The variables acc0 ... acc3 hold output values that are being computed: |
emh203 | 0:3d9c67d97d6f | 108 | * |
emh203 | 0:3d9c67d97d6f | 109 | * 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] |
emh203 | 0:3d9c67d97d6f | 110 | * 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] |
emh203 | 0:3d9c67d97d6f | 111 | * 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] |
emh203 | 0:3d9c67d97d6f | 112 | * 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] |
emh203 | 0:3d9c67d97d6f | 113 | */ |
emh203 | 0:3d9c67d97d6f | 114 | |
emh203 | 0:3d9c67d97d6f | 115 | blkCnt = blockSize >> 2; |
emh203 | 0:3d9c67d97d6f | 116 | |
emh203 | 0:3d9c67d97d6f | 117 | /* First part of the processing with loop unrolling. Compute 4 outputs at a time. |
emh203 | 0:3d9c67d97d6f | 118 | ** a second loop below computes the remaining 1 to 3 samples. */ |
emh203 | 0:3d9c67d97d6f | 119 | while(blkCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 120 | { |
emh203 | 0:3d9c67d97d6f | 121 | /* Copy four new input samples into the state buffer. |
emh203 | 0:3d9c67d97d6f | 122 | ** Use 32-bit SIMD to move the 16-bit data. Only requires two copies. */ |
emh203 | 0:3d9c67d97d6f | 123 | *__SIMD32(pStateCurnt)++ = *__SIMD32(pSrc)++; |
emh203 | 0:3d9c67d97d6f | 124 | *__SIMD32(pStateCurnt)++ = *__SIMD32(pSrc)++; |
emh203 | 0:3d9c67d97d6f | 125 | |
emh203 | 0:3d9c67d97d6f | 126 | /* Set all accumulators to zero */ |
emh203 | 0:3d9c67d97d6f | 127 | acc0 = 0; |
emh203 | 0:3d9c67d97d6f | 128 | acc1 = 0; |
emh203 | 0:3d9c67d97d6f | 129 | acc2 = 0; |
emh203 | 0:3d9c67d97d6f | 130 | acc3 = 0; |
emh203 | 0:3d9c67d97d6f | 131 | |
emh203 | 0:3d9c67d97d6f | 132 | /* Initialize state pointer of type q15 */ |
emh203 | 0:3d9c67d97d6f | 133 | px1 = pState; |
emh203 | 0:3d9c67d97d6f | 134 | |
emh203 | 0:3d9c67d97d6f | 135 | /* Initialize coeff pointer of type q31 */ |
emh203 | 0:3d9c67d97d6f | 136 | pb = pCoeffs; |
emh203 | 0:3d9c67d97d6f | 137 | |
emh203 | 0:3d9c67d97d6f | 138 | /* Read the first two samples from the state buffer: x[n-N], x[n-N-1] */ |
emh203 | 0:3d9c67d97d6f | 139 | x0 = _SIMD32_OFFSET(px1); |
emh203 | 0:3d9c67d97d6f | 140 | |
emh203 | 0:3d9c67d97d6f | 141 | /* Read the third and forth samples from the state buffer: x[n-N-1], x[n-N-2] */ |
emh203 | 0:3d9c67d97d6f | 142 | x1 = _SIMD32_OFFSET(px1 + 1u); |
emh203 | 0:3d9c67d97d6f | 143 | |
emh203 | 0:3d9c67d97d6f | 144 | px1 += 2u; |
emh203 | 0:3d9c67d97d6f | 145 | |
emh203 | 0:3d9c67d97d6f | 146 | /* Loop over the number of taps. Unroll by a factor of 4. |
emh203 | 0:3d9c67d97d6f | 147 | ** Repeat until we've computed numTaps-4 coefficients. */ |
emh203 | 0:3d9c67d97d6f | 148 | tapCnt = numTaps >> 2; |
emh203 | 0:3d9c67d97d6f | 149 | |
emh203 | 0:3d9c67d97d6f | 150 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 151 | { |
emh203 | 0:3d9c67d97d6f | 152 | /* Read the first two coefficients using SIMD: b[N] and b[N-1] coefficients */ |
emh203 | 0:3d9c67d97d6f | 153 | c0 = *__SIMD32(pb)++; |
emh203 | 0:3d9c67d97d6f | 154 | |
emh203 | 0:3d9c67d97d6f | 155 | /* acc0 += b[N] * x[n-N] + b[N-1] * x[n-N-1] */ |
emh203 | 0:3d9c67d97d6f | 156 | acc0 = __SMLALD(x0, c0, acc0); |
emh203 | 0:3d9c67d97d6f | 157 | |
emh203 | 0:3d9c67d97d6f | 158 | /* acc1 += b[N] * x[n-N-1] + b[N-1] * x[n-N-2] */ |
emh203 | 0:3d9c67d97d6f | 159 | acc1 = __SMLALD(x1, c0, acc1); |
emh203 | 0:3d9c67d97d6f | 160 | |
emh203 | 0:3d9c67d97d6f | 161 | /* Read state x[n-N-2], x[n-N-3] */ |
emh203 | 0:3d9c67d97d6f | 162 | x2 = _SIMD32_OFFSET(px1); |
emh203 | 0:3d9c67d97d6f | 163 | |
emh203 | 0:3d9c67d97d6f | 164 | /* Read state x[n-N-3], x[n-N-4] */ |
emh203 | 0:3d9c67d97d6f | 165 | x3 = _SIMD32_OFFSET(px1 + 1u); |
emh203 | 0:3d9c67d97d6f | 166 | |
emh203 | 0:3d9c67d97d6f | 167 | /* acc2 += b[N] * x[n-N-2] + b[N-1] * x[n-N-3] */ |
emh203 | 0:3d9c67d97d6f | 168 | acc2 = __SMLALD(x2, c0, acc2); |
emh203 | 0:3d9c67d97d6f | 169 | |
emh203 | 0:3d9c67d97d6f | 170 | /* acc3 += b[N] * x[n-N-3] + b[N-1] * x[n-N-4] */ |
emh203 | 0:3d9c67d97d6f | 171 | acc3 = __SMLALD(x3, c0, acc3); |
emh203 | 0:3d9c67d97d6f | 172 | |
emh203 | 0:3d9c67d97d6f | 173 | /* Read coefficients b[N-2], b[N-3] */ |
emh203 | 0:3d9c67d97d6f | 174 | c0 = *__SIMD32(pb)++; |
emh203 | 0:3d9c67d97d6f | 175 | |
emh203 | 0:3d9c67d97d6f | 176 | /* acc0 += b[N-2] * x[n-N-2] + b[N-3] * x[n-N-3] */ |
emh203 | 0:3d9c67d97d6f | 177 | acc0 = __SMLALD(x2, c0, acc0); |
emh203 | 0:3d9c67d97d6f | 178 | |
emh203 | 0:3d9c67d97d6f | 179 | /* acc1 += b[N-2] * x[n-N-3] + b[N-3] * x[n-N-4] */ |
emh203 | 0:3d9c67d97d6f | 180 | acc1 = __SMLALD(x3, c0, acc1); |
emh203 | 0:3d9c67d97d6f | 181 | |
emh203 | 0:3d9c67d97d6f | 182 | /* Read state x[n-N-4], x[n-N-5] */ |
emh203 | 0:3d9c67d97d6f | 183 | x0 = _SIMD32_OFFSET(px1 + 2u); |
emh203 | 0:3d9c67d97d6f | 184 | |
emh203 | 0:3d9c67d97d6f | 185 | /* Read state x[n-N-5], x[n-N-6] */ |
emh203 | 0:3d9c67d97d6f | 186 | x1 = _SIMD32_OFFSET(px1 + 3u); |
emh203 | 0:3d9c67d97d6f | 187 | |
emh203 | 0:3d9c67d97d6f | 188 | /* acc2 += b[N-2] * x[n-N-4] + b[N-3] * x[n-N-5] */ |
emh203 | 0:3d9c67d97d6f | 189 | acc2 = __SMLALD(x0, c0, acc2); |
emh203 | 0:3d9c67d97d6f | 190 | |
emh203 | 0:3d9c67d97d6f | 191 | /* acc3 += b[N-2] * x[n-N-5] + b[N-3] * x[n-N-6] */ |
emh203 | 0:3d9c67d97d6f | 192 | acc3 = __SMLALD(x1, c0, acc3); |
emh203 | 0:3d9c67d97d6f | 193 | |
emh203 | 0:3d9c67d97d6f | 194 | px1 += 4u; |
emh203 | 0:3d9c67d97d6f | 195 | |
emh203 | 0:3d9c67d97d6f | 196 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 197 | |
emh203 | 0:3d9c67d97d6f | 198 | } |
emh203 | 0:3d9c67d97d6f | 199 | |
emh203 | 0:3d9c67d97d6f | 200 | |
emh203 | 0:3d9c67d97d6f | 201 | /* If the filter length is not a multiple of 4, compute the remaining filter taps. |
emh203 | 0:3d9c67d97d6f | 202 | ** This is always be 2 taps since the filter length is even. */ |
emh203 | 0:3d9c67d97d6f | 203 | if((numTaps & 0x3u) != 0u) |
emh203 | 0:3d9c67d97d6f | 204 | { |
emh203 | 0:3d9c67d97d6f | 205 | /* Read 2 coefficients */ |
emh203 | 0:3d9c67d97d6f | 206 | c0 = *__SIMD32(pb)++; |
emh203 | 0:3d9c67d97d6f | 207 | |
emh203 | 0:3d9c67d97d6f | 208 | /* Fetch 4 state variables */ |
emh203 | 0:3d9c67d97d6f | 209 | x2 = _SIMD32_OFFSET(px1); |
emh203 | 0:3d9c67d97d6f | 210 | |
emh203 | 0:3d9c67d97d6f | 211 | x3 = _SIMD32_OFFSET(px1 + 1u); |
emh203 | 0:3d9c67d97d6f | 212 | |
emh203 | 0:3d9c67d97d6f | 213 | /* Perform the multiply-accumulates */ |
emh203 | 0:3d9c67d97d6f | 214 | acc0 = __SMLALD(x0, c0, acc0); |
emh203 | 0:3d9c67d97d6f | 215 | |
emh203 | 0:3d9c67d97d6f | 216 | px1 += 2u; |
emh203 | 0:3d9c67d97d6f | 217 | |
emh203 | 0:3d9c67d97d6f | 218 | acc1 = __SMLALD(x1, c0, acc1); |
emh203 | 0:3d9c67d97d6f | 219 | acc2 = __SMLALD(x2, c0, acc2); |
emh203 | 0:3d9c67d97d6f | 220 | acc3 = __SMLALD(x3, c0, acc3); |
emh203 | 0:3d9c67d97d6f | 221 | } |
emh203 | 0:3d9c67d97d6f | 222 | |
emh203 | 0:3d9c67d97d6f | 223 | /* The results in the 4 accumulators are in 2.30 format. Convert to 1.15 with saturation. |
emh203 | 0:3d9c67d97d6f | 224 | ** Then store the 4 outputs in the destination buffer. */ |
emh203 | 0:3d9c67d97d6f | 225 | |
emh203 | 0:3d9c67d97d6f | 226 | #ifndef ARM_MATH_BIG_ENDIAN |
emh203 | 0:3d9c67d97d6f | 227 | |
emh203 | 0:3d9c67d97d6f | 228 | *__SIMD32(pDst)++ = |
emh203 | 0:3d9c67d97d6f | 229 | __PKHBT(__SSAT((acc0 >> 15), 16), __SSAT((acc1 >> 15), 16), 16); |
emh203 | 0:3d9c67d97d6f | 230 | *__SIMD32(pDst)++ = |
emh203 | 0:3d9c67d97d6f | 231 | __PKHBT(__SSAT((acc2 >> 15), 16), __SSAT((acc3 >> 15), 16), 16); |
emh203 | 0:3d9c67d97d6f | 232 | |
emh203 | 0:3d9c67d97d6f | 233 | #else |
emh203 | 0:3d9c67d97d6f | 234 | |
emh203 | 0:3d9c67d97d6f | 235 | *__SIMD32(pDst)++ = |
emh203 | 0:3d9c67d97d6f | 236 | __PKHBT(__SSAT((acc1 >> 15), 16), __SSAT((acc0 >> 15), 16), 16); |
emh203 | 0:3d9c67d97d6f | 237 | *__SIMD32(pDst)++ = |
emh203 | 0:3d9c67d97d6f | 238 | __PKHBT(__SSAT((acc3 >> 15), 16), __SSAT((acc2 >> 15), 16), 16); |
emh203 | 0:3d9c67d97d6f | 239 | |
emh203 | 0:3d9c67d97d6f | 240 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
emh203 | 0:3d9c67d97d6f | 241 | |
emh203 | 0:3d9c67d97d6f | 242 | |
emh203 | 0:3d9c67d97d6f | 243 | |
emh203 | 0:3d9c67d97d6f | 244 | /* Advance the state pointer by 4 to process the next group of 4 samples */ |
emh203 | 0:3d9c67d97d6f | 245 | pState = pState + 4; |
emh203 | 0:3d9c67d97d6f | 246 | |
emh203 | 0:3d9c67d97d6f | 247 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 248 | blkCnt--; |
emh203 | 0:3d9c67d97d6f | 249 | } |
emh203 | 0:3d9c67d97d6f | 250 | |
emh203 | 0:3d9c67d97d6f | 251 | /* If the blockSize is not a multiple of 4, compute any remaining output samples here. |
emh203 | 0:3d9c67d97d6f | 252 | ** No loop unrolling is used. */ |
emh203 | 0:3d9c67d97d6f | 253 | blkCnt = blockSize % 0x4u; |
emh203 | 0:3d9c67d97d6f | 254 | while(blkCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 255 | { |
emh203 | 0:3d9c67d97d6f | 256 | /* Copy two samples into state buffer */ |
emh203 | 0:3d9c67d97d6f | 257 | *pStateCurnt++ = *pSrc++; |
emh203 | 0:3d9c67d97d6f | 258 | |
emh203 | 0:3d9c67d97d6f | 259 | /* Set the accumulator to zero */ |
emh203 | 0:3d9c67d97d6f | 260 | acc0 = 0; |
emh203 | 0:3d9c67d97d6f | 261 | |
emh203 | 0:3d9c67d97d6f | 262 | /* Initialize state pointer of type q15 */ |
emh203 | 0:3d9c67d97d6f | 263 | px1 = pState; |
emh203 | 0:3d9c67d97d6f | 264 | |
emh203 | 0:3d9c67d97d6f | 265 | /* Initialize coeff pointer of type q31 */ |
emh203 | 0:3d9c67d97d6f | 266 | pb = pCoeffs; |
emh203 | 0:3d9c67d97d6f | 267 | |
emh203 | 0:3d9c67d97d6f | 268 | tapCnt = numTaps >> 1; |
emh203 | 0:3d9c67d97d6f | 269 | |
emh203 | 0:3d9c67d97d6f | 270 | do |
emh203 | 0:3d9c67d97d6f | 271 | { |
emh203 | 0:3d9c67d97d6f | 272 | |
emh203 | 0:3d9c67d97d6f | 273 | c0 = *__SIMD32(pb)++; |
emh203 | 0:3d9c67d97d6f | 274 | x0 = *__SIMD32(px1)++; |
emh203 | 0:3d9c67d97d6f | 275 | |
emh203 | 0:3d9c67d97d6f | 276 | acc0 = __SMLALD(x0, c0, acc0); |
emh203 | 0:3d9c67d97d6f | 277 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 278 | } |
emh203 | 0:3d9c67d97d6f | 279 | while(tapCnt > 0u); |
emh203 | 0:3d9c67d97d6f | 280 | |
emh203 | 0:3d9c67d97d6f | 281 | /* The result is in 2.30 format. Convert to 1.15 with saturation. |
emh203 | 0:3d9c67d97d6f | 282 | ** Then store the output in the destination buffer. */ |
emh203 | 0:3d9c67d97d6f | 283 | *pDst++ = (q15_t) (__SSAT((acc0 >> 15), 16)); |
emh203 | 0:3d9c67d97d6f | 284 | |
emh203 | 0:3d9c67d97d6f | 285 | /* Advance state pointer by 1 for the next sample */ |
emh203 | 0:3d9c67d97d6f | 286 | pState = pState + 1; |
emh203 | 0:3d9c67d97d6f | 287 | |
emh203 | 0:3d9c67d97d6f | 288 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 289 | blkCnt--; |
emh203 | 0:3d9c67d97d6f | 290 | } |
emh203 | 0:3d9c67d97d6f | 291 | |
emh203 | 0:3d9c67d97d6f | 292 | /* Processing is complete. |
emh203 | 0:3d9c67d97d6f | 293 | ** Now copy the last numTaps - 1 samples to the satrt of the state buffer. |
emh203 | 0:3d9c67d97d6f | 294 | ** This prepares the state buffer for the next function call. */ |
emh203 | 0:3d9c67d97d6f | 295 | |
emh203 | 0:3d9c67d97d6f | 296 | /* Points to the start of the state buffer */ |
emh203 | 0:3d9c67d97d6f | 297 | pStateCurnt = S->pState; |
emh203 | 0:3d9c67d97d6f | 298 | |
emh203 | 0:3d9c67d97d6f | 299 | /* Calculation of count for copying integer writes */ |
emh203 | 0:3d9c67d97d6f | 300 | tapCnt = (numTaps - 1u) >> 2; |
emh203 | 0:3d9c67d97d6f | 301 | |
emh203 | 0:3d9c67d97d6f | 302 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 303 | { |
emh203 | 0:3d9c67d97d6f | 304 | |
emh203 | 0:3d9c67d97d6f | 305 | /* Copy state values to start of state buffer */ |
emh203 | 0:3d9c67d97d6f | 306 | *__SIMD32(pStateCurnt)++ = *__SIMD32(pState)++; |
emh203 | 0:3d9c67d97d6f | 307 | *__SIMD32(pStateCurnt)++ = *__SIMD32(pState)++; |
emh203 | 0:3d9c67d97d6f | 308 | |
emh203 | 0:3d9c67d97d6f | 309 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 310 | |
emh203 | 0:3d9c67d97d6f | 311 | } |
emh203 | 0:3d9c67d97d6f | 312 | |
emh203 | 0:3d9c67d97d6f | 313 | /* Calculation of count for remaining q15_t data */ |
emh203 | 0:3d9c67d97d6f | 314 | tapCnt = (numTaps - 1u) % 0x4u; |
emh203 | 0:3d9c67d97d6f | 315 | |
emh203 | 0:3d9c67d97d6f | 316 | /* copy remaining data */ |
emh203 | 0:3d9c67d97d6f | 317 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 318 | { |
emh203 | 0:3d9c67d97d6f | 319 | *pStateCurnt++ = *pState++; |
emh203 | 0:3d9c67d97d6f | 320 | |
emh203 | 0:3d9c67d97d6f | 321 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 322 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 323 | } |
emh203 | 0:3d9c67d97d6f | 324 | } |
emh203 | 0:3d9c67d97d6f | 325 | |
emh203 | 0:3d9c67d97d6f | 326 | #else /* UNALIGNED_SUPPORT_DISABLE */ |
emh203 | 0:3d9c67d97d6f | 327 | |
emh203 | 0:3d9c67d97d6f | 328 | void arm_fir_q15( |
emh203 | 0:3d9c67d97d6f | 329 | const arm_fir_instance_q15 * S, |
emh203 | 0:3d9c67d97d6f | 330 | q15_t * pSrc, |
emh203 | 0:3d9c67d97d6f | 331 | q15_t * pDst, |
emh203 | 0:3d9c67d97d6f | 332 | uint32_t blockSize) |
emh203 | 0:3d9c67d97d6f | 333 | { |
emh203 | 0:3d9c67d97d6f | 334 | q15_t *pState = S->pState; /* State pointer */ |
emh203 | 0:3d9c67d97d6f | 335 | q15_t *pCoeffs = S->pCoeffs; /* Coefficient pointer */ |
emh203 | 0:3d9c67d97d6f | 336 | q15_t *pStateCurnt; /* Points to the current sample of the state */ |
emh203 | 0:3d9c67d97d6f | 337 | q63_t acc0, acc1, acc2, acc3; /* Accumulators */ |
emh203 | 0:3d9c67d97d6f | 338 | q15_t *pb; /* Temporary pointer for coefficient buffer */ |
emh203 | 0:3d9c67d97d6f | 339 | q15_t *px; /* Temporary q31 pointer for SIMD state buffer accesses */ |
emh203 | 0:3d9c67d97d6f | 340 | q31_t x0, x1, x2, c0; /* Temporary variables to hold SIMD state and coefficient values */ |
emh203 | 0:3d9c67d97d6f | 341 | uint32_t numTaps = S->numTaps; /* Number of taps in the filter */ |
emh203 | 0:3d9c67d97d6f | 342 | uint32_t tapCnt, blkCnt; /* Loop counters */ |
emh203 | 0:3d9c67d97d6f | 343 | |
emh203 | 0:3d9c67d97d6f | 344 | |
emh203 | 0:3d9c67d97d6f | 345 | /* S->pState points to state array which contains previous frame (numTaps - 1) samples */ |
emh203 | 0:3d9c67d97d6f | 346 | /* pStateCurnt points to the location where the new input data should be written */ |
emh203 | 0:3d9c67d97d6f | 347 | pStateCurnt = &(S->pState[(numTaps - 1u)]); |
emh203 | 0:3d9c67d97d6f | 348 | |
emh203 | 0:3d9c67d97d6f | 349 | /* Apply loop unrolling and compute 4 output values simultaneously. |
emh203 | 0:3d9c67d97d6f | 350 | * The variables acc0 ... acc3 hold output values that are being computed: |
emh203 | 0:3d9c67d97d6f | 351 | * |
emh203 | 0:3d9c67d97d6f | 352 | * 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] |
emh203 | 0:3d9c67d97d6f | 353 | * 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] |
emh203 | 0:3d9c67d97d6f | 354 | * 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] |
emh203 | 0:3d9c67d97d6f | 355 | * 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] |
emh203 | 0:3d9c67d97d6f | 356 | */ |
emh203 | 0:3d9c67d97d6f | 357 | |
emh203 | 0:3d9c67d97d6f | 358 | blkCnt = blockSize >> 2; |
emh203 | 0:3d9c67d97d6f | 359 | |
emh203 | 0:3d9c67d97d6f | 360 | /* First part of the processing with loop unrolling. Compute 4 outputs at a time. |
emh203 | 0:3d9c67d97d6f | 361 | ** a second loop below computes the remaining 1 to 3 samples. */ |
emh203 | 0:3d9c67d97d6f | 362 | while(blkCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 363 | { |
emh203 | 0:3d9c67d97d6f | 364 | /* Copy four new input samples into the state buffer. |
emh203 | 0:3d9c67d97d6f | 365 | ** Use 32-bit SIMD to move the 16-bit data. Only requires two copies. */ |
emh203 | 0:3d9c67d97d6f | 366 | *pStateCurnt++ = *pSrc++; |
emh203 | 0:3d9c67d97d6f | 367 | *pStateCurnt++ = *pSrc++; |
emh203 | 0:3d9c67d97d6f | 368 | *pStateCurnt++ = *pSrc++; |
emh203 | 0:3d9c67d97d6f | 369 | *pStateCurnt++ = *pSrc++; |
emh203 | 0:3d9c67d97d6f | 370 | |
emh203 | 0:3d9c67d97d6f | 371 | |
emh203 | 0:3d9c67d97d6f | 372 | /* Set all accumulators to zero */ |
emh203 | 0:3d9c67d97d6f | 373 | acc0 = 0; |
emh203 | 0:3d9c67d97d6f | 374 | acc1 = 0; |
emh203 | 0:3d9c67d97d6f | 375 | acc2 = 0; |
emh203 | 0:3d9c67d97d6f | 376 | acc3 = 0; |
emh203 | 0:3d9c67d97d6f | 377 | |
emh203 | 0:3d9c67d97d6f | 378 | /* Typecast q15_t pointer to q31_t pointer for state reading in q31_t */ |
emh203 | 0:3d9c67d97d6f | 379 | px = pState; |
emh203 | 0:3d9c67d97d6f | 380 | |
emh203 | 0:3d9c67d97d6f | 381 | /* Typecast q15_t pointer to q31_t pointer for coefficient reading in q31_t */ |
emh203 | 0:3d9c67d97d6f | 382 | pb = pCoeffs; |
emh203 | 0:3d9c67d97d6f | 383 | |
emh203 | 0:3d9c67d97d6f | 384 | /* Read the first two samples from the state buffer: x[n-N], x[n-N-1] */ |
emh203 | 0:3d9c67d97d6f | 385 | x0 = *__SIMD32(px)++; |
emh203 | 0:3d9c67d97d6f | 386 | |
emh203 | 0:3d9c67d97d6f | 387 | /* Read the third and forth samples from the state buffer: x[n-N-2], x[n-N-3] */ |
emh203 | 0:3d9c67d97d6f | 388 | x2 = *__SIMD32(px)++; |
emh203 | 0:3d9c67d97d6f | 389 | |
emh203 | 0:3d9c67d97d6f | 390 | /* Loop over the number of taps. Unroll by a factor of 4. |
emh203 | 0:3d9c67d97d6f | 391 | ** Repeat until we've computed numTaps-(numTaps%4) coefficients. */ |
emh203 | 0:3d9c67d97d6f | 392 | tapCnt = numTaps >> 2; |
emh203 | 0:3d9c67d97d6f | 393 | |
emh203 | 0:3d9c67d97d6f | 394 | while(tapCnt > 0) |
emh203 | 0:3d9c67d97d6f | 395 | { |
emh203 | 0:3d9c67d97d6f | 396 | /* Read the first two coefficients using SIMD: b[N] and b[N-1] coefficients */ |
emh203 | 0:3d9c67d97d6f | 397 | c0 = *__SIMD32(pb)++; |
emh203 | 0:3d9c67d97d6f | 398 | |
emh203 | 0:3d9c67d97d6f | 399 | /* acc0 += b[N] * x[n-N] + b[N-1] * x[n-N-1] */ |
emh203 | 0:3d9c67d97d6f | 400 | acc0 = __SMLALD(x0, c0, acc0); |
emh203 | 0:3d9c67d97d6f | 401 | |
emh203 | 0:3d9c67d97d6f | 402 | /* acc2 += b[N] * x[n-N-2] + b[N-1] * x[n-N-3] */ |
emh203 | 0:3d9c67d97d6f | 403 | acc2 = __SMLALD(x2, c0, acc2); |
emh203 | 0:3d9c67d97d6f | 404 | |
emh203 | 0:3d9c67d97d6f | 405 | /* pack x[n-N-1] and x[n-N-2] */ |
emh203 | 0:3d9c67d97d6f | 406 | #ifndef ARM_MATH_BIG_ENDIAN |
emh203 | 0:3d9c67d97d6f | 407 | x1 = __PKHBT(x2, x0, 0); |
emh203 | 0:3d9c67d97d6f | 408 | #else |
emh203 | 0:3d9c67d97d6f | 409 | x1 = __PKHBT(x0, x2, 0); |
emh203 | 0:3d9c67d97d6f | 410 | #endif |
emh203 | 0:3d9c67d97d6f | 411 | |
emh203 | 0:3d9c67d97d6f | 412 | /* Read state x[n-N-4], x[n-N-5] */ |
emh203 | 0:3d9c67d97d6f | 413 | x0 = _SIMD32_OFFSET(px); |
emh203 | 0:3d9c67d97d6f | 414 | |
emh203 | 0:3d9c67d97d6f | 415 | /* acc1 += b[N] * x[n-N-1] + b[N-1] * x[n-N-2] */ |
emh203 | 0:3d9c67d97d6f | 416 | acc1 = __SMLALDX(x1, c0, acc1); |
emh203 | 0:3d9c67d97d6f | 417 | |
emh203 | 0:3d9c67d97d6f | 418 | /* pack x[n-N-3] and x[n-N-4] */ |
emh203 | 0:3d9c67d97d6f | 419 | #ifndef ARM_MATH_BIG_ENDIAN |
emh203 | 0:3d9c67d97d6f | 420 | x1 = __PKHBT(x0, x2, 0); |
emh203 | 0:3d9c67d97d6f | 421 | #else |
emh203 | 0:3d9c67d97d6f | 422 | x1 = __PKHBT(x2, x0, 0); |
emh203 | 0:3d9c67d97d6f | 423 | #endif |
emh203 | 0:3d9c67d97d6f | 424 | |
emh203 | 0:3d9c67d97d6f | 425 | /* acc3 += b[N] * x[n-N-3] + b[N-1] * x[n-N-4] */ |
emh203 | 0:3d9c67d97d6f | 426 | acc3 = __SMLALDX(x1, c0, acc3); |
emh203 | 0:3d9c67d97d6f | 427 | |
emh203 | 0:3d9c67d97d6f | 428 | /* Read coefficients b[N-2], b[N-3] */ |
emh203 | 0:3d9c67d97d6f | 429 | c0 = *__SIMD32(pb)++; |
emh203 | 0:3d9c67d97d6f | 430 | |
emh203 | 0:3d9c67d97d6f | 431 | /* acc0 += b[N-2] * x[n-N-2] + b[N-3] * x[n-N-3] */ |
emh203 | 0:3d9c67d97d6f | 432 | acc0 = __SMLALD(x2, c0, acc0); |
emh203 | 0:3d9c67d97d6f | 433 | |
emh203 | 0:3d9c67d97d6f | 434 | /* Read state x[n-N-6], x[n-N-7] with offset */ |
emh203 | 0:3d9c67d97d6f | 435 | x2 = _SIMD32_OFFSET(px + 2u); |
emh203 | 0:3d9c67d97d6f | 436 | |
emh203 | 0:3d9c67d97d6f | 437 | /* acc2 += b[N-2] * x[n-N-4] + b[N-3] * x[n-N-5] */ |
emh203 | 0:3d9c67d97d6f | 438 | acc2 = __SMLALD(x0, c0, acc2); |
emh203 | 0:3d9c67d97d6f | 439 | |
emh203 | 0:3d9c67d97d6f | 440 | /* acc1 += b[N-2] * x[n-N-3] + b[N-3] * x[n-N-4] */ |
emh203 | 0:3d9c67d97d6f | 441 | acc1 = __SMLALDX(x1, c0, acc1); |
emh203 | 0:3d9c67d97d6f | 442 | |
emh203 | 0:3d9c67d97d6f | 443 | /* pack x[n-N-5] and x[n-N-6] */ |
emh203 | 0:3d9c67d97d6f | 444 | #ifndef ARM_MATH_BIG_ENDIAN |
emh203 | 0:3d9c67d97d6f | 445 | x1 = __PKHBT(x2, x0, 0); |
emh203 | 0:3d9c67d97d6f | 446 | #else |
emh203 | 0:3d9c67d97d6f | 447 | x1 = __PKHBT(x0, x2, 0); |
emh203 | 0:3d9c67d97d6f | 448 | #endif |
emh203 | 0:3d9c67d97d6f | 449 | |
emh203 | 0:3d9c67d97d6f | 450 | /* acc3 += b[N-2] * x[n-N-5] + b[N-3] * x[n-N-6] */ |
emh203 | 0:3d9c67d97d6f | 451 | acc3 = __SMLALDX(x1, c0, acc3); |
emh203 | 0:3d9c67d97d6f | 452 | |
emh203 | 0:3d9c67d97d6f | 453 | /* Update state pointer for next state reading */ |
emh203 | 0:3d9c67d97d6f | 454 | px += 4u; |
emh203 | 0:3d9c67d97d6f | 455 | |
emh203 | 0:3d9c67d97d6f | 456 | /* Decrement tap count */ |
emh203 | 0:3d9c67d97d6f | 457 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 458 | |
emh203 | 0:3d9c67d97d6f | 459 | } |
emh203 | 0:3d9c67d97d6f | 460 | |
emh203 | 0:3d9c67d97d6f | 461 | /* If the filter length is not a multiple of 4, compute the remaining filter taps. |
emh203 | 0:3d9c67d97d6f | 462 | ** This is always be 2 taps since the filter length is even. */ |
emh203 | 0:3d9c67d97d6f | 463 | if((numTaps & 0x3u) != 0u) |
emh203 | 0:3d9c67d97d6f | 464 | { |
emh203 | 0:3d9c67d97d6f | 465 | |
emh203 | 0:3d9c67d97d6f | 466 | /* Read last two coefficients */ |
emh203 | 0:3d9c67d97d6f | 467 | c0 = *__SIMD32(pb)++; |
emh203 | 0:3d9c67d97d6f | 468 | |
emh203 | 0:3d9c67d97d6f | 469 | /* Perform the multiply-accumulates */ |
emh203 | 0:3d9c67d97d6f | 470 | acc0 = __SMLALD(x0, c0, acc0); |
emh203 | 0:3d9c67d97d6f | 471 | acc2 = __SMLALD(x2, c0, acc2); |
emh203 | 0:3d9c67d97d6f | 472 | |
emh203 | 0:3d9c67d97d6f | 473 | /* pack state variables */ |
emh203 | 0:3d9c67d97d6f | 474 | #ifndef ARM_MATH_BIG_ENDIAN |
emh203 | 0:3d9c67d97d6f | 475 | x1 = __PKHBT(x2, x0, 0); |
emh203 | 0:3d9c67d97d6f | 476 | #else |
emh203 | 0:3d9c67d97d6f | 477 | x1 = __PKHBT(x0, x2, 0); |
emh203 | 0:3d9c67d97d6f | 478 | #endif |
emh203 | 0:3d9c67d97d6f | 479 | |
emh203 | 0:3d9c67d97d6f | 480 | /* Read last state variables */ |
emh203 | 0:3d9c67d97d6f | 481 | x0 = *__SIMD32(px); |
emh203 | 0:3d9c67d97d6f | 482 | |
emh203 | 0:3d9c67d97d6f | 483 | /* Perform the multiply-accumulates */ |
emh203 | 0:3d9c67d97d6f | 484 | acc1 = __SMLALDX(x1, c0, acc1); |
emh203 | 0:3d9c67d97d6f | 485 | |
emh203 | 0:3d9c67d97d6f | 486 | /* pack state variables */ |
emh203 | 0:3d9c67d97d6f | 487 | #ifndef ARM_MATH_BIG_ENDIAN |
emh203 | 0:3d9c67d97d6f | 488 | x1 = __PKHBT(x0, x2, 0); |
emh203 | 0:3d9c67d97d6f | 489 | #else |
emh203 | 0:3d9c67d97d6f | 490 | x1 = __PKHBT(x2, x0, 0); |
emh203 | 0:3d9c67d97d6f | 491 | #endif |
emh203 | 0:3d9c67d97d6f | 492 | |
emh203 | 0:3d9c67d97d6f | 493 | /* Perform the multiply-accumulates */ |
emh203 | 0:3d9c67d97d6f | 494 | acc3 = __SMLALDX(x1, c0, acc3); |
emh203 | 0:3d9c67d97d6f | 495 | } |
emh203 | 0:3d9c67d97d6f | 496 | |
emh203 | 0:3d9c67d97d6f | 497 | /* The results in the 4 accumulators are in 2.30 format. Convert to 1.15 with saturation. |
emh203 | 0:3d9c67d97d6f | 498 | ** Then store the 4 outputs in the destination buffer. */ |
emh203 | 0:3d9c67d97d6f | 499 | |
emh203 | 0:3d9c67d97d6f | 500 | #ifndef ARM_MATH_BIG_ENDIAN |
emh203 | 0:3d9c67d97d6f | 501 | |
emh203 | 0:3d9c67d97d6f | 502 | *__SIMD32(pDst)++ = |
emh203 | 0:3d9c67d97d6f | 503 | __PKHBT(__SSAT((acc0 >> 15), 16), __SSAT((acc1 >> 15), 16), 16); |
emh203 | 0:3d9c67d97d6f | 504 | |
emh203 | 0:3d9c67d97d6f | 505 | *__SIMD32(pDst)++ = |
emh203 | 0:3d9c67d97d6f | 506 | __PKHBT(__SSAT((acc2 >> 15), 16), __SSAT((acc3 >> 15), 16), 16); |
emh203 | 0:3d9c67d97d6f | 507 | |
emh203 | 0:3d9c67d97d6f | 508 | #else |
emh203 | 0:3d9c67d97d6f | 509 | |
emh203 | 0:3d9c67d97d6f | 510 | *__SIMD32(pDst)++ = |
emh203 | 0:3d9c67d97d6f | 511 | __PKHBT(__SSAT((acc1 >> 15), 16), __SSAT((acc0 >> 15), 16), 16); |
emh203 | 0:3d9c67d97d6f | 512 | |
emh203 | 0:3d9c67d97d6f | 513 | *__SIMD32(pDst)++ = |
emh203 | 0:3d9c67d97d6f | 514 | __PKHBT(__SSAT((acc3 >> 15), 16), __SSAT((acc2 >> 15), 16), 16); |
emh203 | 0:3d9c67d97d6f | 515 | |
emh203 | 0:3d9c67d97d6f | 516 | #endif /* #ifndef ARM_MATH_BIG_ENDIAN */ |
emh203 | 0:3d9c67d97d6f | 517 | |
emh203 | 0:3d9c67d97d6f | 518 | /* Advance the state pointer by 4 to process the next group of 4 samples */ |
emh203 | 0:3d9c67d97d6f | 519 | pState = pState + 4; |
emh203 | 0:3d9c67d97d6f | 520 | |
emh203 | 0:3d9c67d97d6f | 521 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 522 | blkCnt--; |
emh203 | 0:3d9c67d97d6f | 523 | } |
emh203 | 0:3d9c67d97d6f | 524 | |
emh203 | 0:3d9c67d97d6f | 525 | /* If the blockSize is not a multiple of 4, compute any remaining output samples here. |
emh203 | 0:3d9c67d97d6f | 526 | ** No loop unrolling is used. */ |
emh203 | 0:3d9c67d97d6f | 527 | blkCnt = blockSize % 0x4u; |
emh203 | 0:3d9c67d97d6f | 528 | while(blkCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 529 | { |
emh203 | 0:3d9c67d97d6f | 530 | /* Copy two samples into state buffer */ |
emh203 | 0:3d9c67d97d6f | 531 | *pStateCurnt++ = *pSrc++; |
emh203 | 0:3d9c67d97d6f | 532 | |
emh203 | 0:3d9c67d97d6f | 533 | /* Set the accumulator to zero */ |
emh203 | 0:3d9c67d97d6f | 534 | acc0 = 0; |
emh203 | 0:3d9c67d97d6f | 535 | |
emh203 | 0:3d9c67d97d6f | 536 | /* Use SIMD to hold states and coefficients */ |
emh203 | 0:3d9c67d97d6f | 537 | px = pState; |
emh203 | 0:3d9c67d97d6f | 538 | pb = pCoeffs; |
emh203 | 0:3d9c67d97d6f | 539 | |
emh203 | 0:3d9c67d97d6f | 540 | tapCnt = numTaps >> 1u; |
emh203 | 0:3d9c67d97d6f | 541 | |
emh203 | 0:3d9c67d97d6f | 542 | do |
emh203 | 0:3d9c67d97d6f | 543 | { |
emh203 | 0:3d9c67d97d6f | 544 | acc0 += (q31_t) * px++ * *pb++; |
emh203 | 0:3d9c67d97d6f | 545 | acc0 += (q31_t) * px++ * *pb++; |
emh203 | 0:3d9c67d97d6f | 546 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 547 | } |
emh203 | 0:3d9c67d97d6f | 548 | while(tapCnt > 0u); |
emh203 | 0:3d9c67d97d6f | 549 | |
emh203 | 0:3d9c67d97d6f | 550 | /* The result is in 2.30 format. Convert to 1.15 with saturation. |
emh203 | 0:3d9c67d97d6f | 551 | ** Then store the output in the destination buffer. */ |
emh203 | 0:3d9c67d97d6f | 552 | *pDst++ = (q15_t) (__SSAT((acc0 >> 15), 16)); |
emh203 | 0:3d9c67d97d6f | 553 | |
emh203 | 0:3d9c67d97d6f | 554 | /* Advance state pointer by 1 for the next sample */ |
emh203 | 0:3d9c67d97d6f | 555 | pState = pState + 1u; |
emh203 | 0:3d9c67d97d6f | 556 | |
emh203 | 0:3d9c67d97d6f | 557 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 558 | blkCnt--; |
emh203 | 0:3d9c67d97d6f | 559 | } |
emh203 | 0:3d9c67d97d6f | 560 | |
emh203 | 0:3d9c67d97d6f | 561 | /* Processing is complete. |
emh203 | 0:3d9c67d97d6f | 562 | ** Now copy the last numTaps - 1 samples to the satrt of the state buffer. |
emh203 | 0:3d9c67d97d6f | 563 | ** This prepares the state buffer for the next function call. */ |
emh203 | 0:3d9c67d97d6f | 564 | |
emh203 | 0:3d9c67d97d6f | 565 | /* Points to the start of the state buffer */ |
emh203 | 0:3d9c67d97d6f | 566 | pStateCurnt = S->pState; |
emh203 | 0:3d9c67d97d6f | 567 | |
emh203 | 0:3d9c67d97d6f | 568 | /* Calculation of count for copying integer writes */ |
emh203 | 0:3d9c67d97d6f | 569 | tapCnt = (numTaps - 1u) >> 2; |
emh203 | 0:3d9c67d97d6f | 570 | |
emh203 | 0:3d9c67d97d6f | 571 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 572 | { |
emh203 | 0:3d9c67d97d6f | 573 | *pStateCurnt++ = *pState++; |
emh203 | 0:3d9c67d97d6f | 574 | *pStateCurnt++ = *pState++; |
emh203 | 0:3d9c67d97d6f | 575 | *pStateCurnt++ = *pState++; |
emh203 | 0:3d9c67d97d6f | 576 | *pStateCurnt++ = *pState++; |
emh203 | 0:3d9c67d97d6f | 577 | |
emh203 | 0:3d9c67d97d6f | 578 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 579 | |
emh203 | 0:3d9c67d97d6f | 580 | } |
emh203 | 0:3d9c67d97d6f | 581 | |
emh203 | 0:3d9c67d97d6f | 582 | /* Calculation of count for remaining q15_t data */ |
emh203 | 0:3d9c67d97d6f | 583 | tapCnt = (numTaps - 1u) % 0x4u; |
emh203 | 0:3d9c67d97d6f | 584 | |
emh203 | 0:3d9c67d97d6f | 585 | /* copy remaining data */ |
emh203 | 0:3d9c67d97d6f | 586 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 587 | { |
emh203 | 0:3d9c67d97d6f | 588 | *pStateCurnt++ = *pState++; |
emh203 | 0:3d9c67d97d6f | 589 | |
emh203 | 0:3d9c67d97d6f | 590 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 591 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 592 | } |
emh203 | 0:3d9c67d97d6f | 593 | } |
emh203 | 0:3d9c67d97d6f | 594 | |
emh203 | 0:3d9c67d97d6f | 595 | |
emh203 | 0:3d9c67d97d6f | 596 | #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */ |
emh203 | 0:3d9c67d97d6f | 597 | |
emh203 | 0:3d9c67d97d6f | 598 | #else /* ARM_MATH_CM0_FAMILY */ |
emh203 | 0:3d9c67d97d6f | 599 | |
emh203 | 0:3d9c67d97d6f | 600 | |
emh203 | 0:3d9c67d97d6f | 601 | /* Run the below code for Cortex-M0 */ |
emh203 | 0:3d9c67d97d6f | 602 | |
emh203 | 0:3d9c67d97d6f | 603 | void arm_fir_q15( |
emh203 | 0:3d9c67d97d6f | 604 | const arm_fir_instance_q15 * S, |
emh203 | 0:3d9c67d97d6f | 605 | q15_t * pSrc, |
emh203 | 0:3d9c67d97d6f | 606 | q15_t * pDst, |
emh203 | 0:3d9c67d97d6f | 607 | uint32_t blockSize) |
emh203 | 0:3d9c67d97d6f | 608 | { |
emh203 | 0:3d9c67d97d6f | 609 | q15_t *pState = S->pState; /* State pointer */ |
emh203 | 0:3d9c67d97d6f | 610 | q15_t *pCoeffs = S->pCoeffs; /* Coefficient pointer */ |
emh203 | 0:3d9c67d97d6f | 611 | q15_t *pStateCurnt; /* Points to the current sample of the state */ |
emh203 | 0:3d9c67d97d6f | 612 | |
emh203 | 0:3d9c67d97d6f | 613 | |
emh203 | 0:3d9c67d97d6f | 614 | |
emh203 | 0:3d9c67d97d6f | 615 | q15_t *px; /* Temporary pointer for state buffer */ |
emh203 | 0:3d9c67d97d6f | 616 | q15_t *pb; /* Temporary pointer for coefficient buffer */ |
emh203 | 0:3d9c67d97d6f | 617 | q63_t acc; /* Accumulator */ |
emh203 | 0:3d9c67d97d6f | 618 | uint32_t numTaps = S->numTaps; /* Number of nTaps in the filter */ |
emh203 | 0:3d9c67d97d6f | 619 | uint32_t tapCnt, blkCnt; /* Loop counters */ |
emh203 | 0:3d9c67d97d6f | 620 | |
emh203 | 0:3d9c67d97d6f | 621 | /* S->pState buffer contains previous frame (numTaps - 1) samples */ |
emh203 | 0:3d9c67d97d6f | 622 | /* pStateCurnt points to the location where the new input data should be written */ |
emh203 | 0:3d9c67d97d6f | 623 | pStateCurnt = &(S->pState[(numTaps - 1u)]); |
emh203 | 0:3d9c67d97d6f | 624 | |
emh203 | 0:3d9c67d97d6f | 625 | /* Initialize blkCnt with blockSize */ |
emh203 | 0:3d9c67d97d6f | 626 | blkCnt = blockSize; |
emh203 | 0:3d9c67d97d6f | 627 | |
emh203 | 0:3d9c67d97d6f | 628 | while(blkCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 629 | { |
emh203 | 0:3d9c67d97d6f | 630 | /* Copy one sample at a time into state buffer */ |
emh203 | 0:3d9c67d97d6f | 631 | *pStateCurnt++ = *pSrc++; |
emh203 | 0:3d9c67d97d6f | 632 | |
emh203 | 0:3d9c67d97d6f | 633 | /* Set the accumulator to zero */ |
emh203 | 0:3d9c67d97d6f | 634 | acc = 0; |
emh203 | 0:3d9c67d97d6f | 635 | |
emh203 | 0:3d9c67d97d6f | 636 | /* Initialize state pointer */ |
emh203 | 0:3d9c67d97d6f | 637 | px = pState; |
emh203 | 0:3d9c67d97d6f | 638 | |
emh203 | 0:3d9c67d97d6f | 639 | /* Initialize Coefficient pointer */ |
emh203 | 0:3d9c67d97d6f | 640 | pb = pCoeffs; |
emh203 | 0:3d9c67d97d6f | 641 | |
emh203 | 0:3d9c67d97d6f | 642 | tapCnt = numTaps; |
emh203 | 0:3d9c67d97d6f | 643 | |
emh203 | 0:3d9c67d97d6f | 644 | /* Perform the multiply-accumulates */ |
emh203 | 0:3d9c67d97d6f | 645 | do |
emh203 | 0:3d9c67d97d6f | 646 | { |
emh203 | 0:3d9c67d97d6f | 647 | /* acc = 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] */ |
emh203 | 0:3d9c67d97d6f | 648 | acc += (q31_t) * px++ * *pb++; |
emh203 | 0:3d9c67d97d6f | 649 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 650 | } while(tapCnt > 0u); |
emh203 | 0:3d9c67d97d6f | 651 | |
emh203 | 0:3d9c67d97d6f | 652 | /* The result is in 2.30 format. Convert to 1.15 |
emh203 | 0:3d9c67d97d6f | 653 | ** Then store the output in the destination buffer. */ |
emh203 | 0:3d9c67d97d6f | 654 | *pDst++ = (q15_t) __SSAT((acc >> 15u), 16); |
emh203 | 0:3d9c67d97d6f | 655 | |
emh203 | 0:3d9c67d97d6f | 656 | /* Advance state pointer by 1 for the next sample */ |
emh203 | 0:3d9c67d97d6f | 657 | pState = pState + 1; |
emh203 | 0:3d9c67d97d6f | 658 | |
emh203 | 0:3d9c67d97d6f | 659 | /* Decrement the samples loop counter */ |
emh203 | 0:3d9c67d97d6f | 660 | blkCnt--; |
emh203 | 0:3d9c67d97d6f | 661 | } |
emh203 | 0:3d9c67d97d6f | 662 | |
emh203 | 0:3d9c67d97d6f | 663 | /* Processing is complete. |
emh203 | 0:3d9c67d97d6f | 664 | ** Now copy the last numTaps - 1 samples to the satrt of the state buffer. |
emh203 | 0:3d9c67d97d6f | 665 | ** This prepares the state buffer for the next function call. */ |
emh203 | 0:3d9c67d97d6f | 666 | |
emh203 | 0:3d9c67d97d6f | 667 | /* Points to the start of the state buffer */ |
emh203 | 0:3d9c67d97d6f | 668 | pStateCurnt = S->pState; |
emh203 | 0:3d9c67d97d6f | 669 | |
emh203 | 0:3d9c67d97d6f | 670 | /* Copy numTaps number of values */ |
emh203 | 0:3d9c67d97d6f | 671 | tapCnt = (numTaps - 1u); |
emh203 | 0:3d9c67d97d6f | 672 | |
emh203 | 0:3d9c67d97d6f | 673 | /* copy data */ |
emh203 | 0:3d9c67d97d6f | 674 | while(tapCnt > 0u) |
emh203 | 0:3d9c67d97d6f | 675 | { |
emh203 | 0:3d9c67d97d6f | 676 | *pStateCurnt++ = *pState++; |
emh203 | 0:3d9c67d97d6f | 677 | |
emh203 | 0:3d9c67d97d6f | 678 | /* Decrement the loop counter */ |
emh203 | 0:3d9c67d97d6f | 679 | tapCnt--; |
emh203 | 0:3d9c67d97d6f | 680 | } |
emh203 | 0:3d9c67d97d6f | 681 | |
emh203 | 0:3d9c67d97d6f | 682 | } |
emh203 | 0:3d9c67d97d6f | 683 | |
emh203 | 0:3d9c67d97d6f | 684 | #endif /* #ifndef ARM_MATH_CM0_FAMILY */ |
emh203 | 0:3d9c67d97d6f | 685 | |
emh203 | 0:3d9c67d97d6f | 686 | |
emh203 | 0:3d9c67d97d6f | 687 | |
emh203 | 0:3d9c67d97d6f | 688 | |
emh203 | 0:3d9c67d97d6f | 689 | /** |
emh203 | 0:3d9c67d97d6f | 690 | * @} end of FIR group |
emh203 | 0:3d9c67d97d6f | 691 | */ |