| /* ---------------------------------------------------------------------- |
| * Project: CMSIS DSP Library |
| * Title: arm_correlate_opt_q15.c |
| * Description: Correlation of Q15 sequences |
| * |
| * $Date: 18. March 2019 |
| * $Revision: V1.6.0 |
| * |
| * Target Processor: Cortex-M cores |
| * -------------------------------------------------------------------- */ |
| /* |
| * Copyright (C) 2010-2019 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 Corr |
| @{ |
| */ |
| |
| /** |
| @brief Correlation of Q15 sequences. |
| @param[in] pSrcA points to the first input sequence |
| @param[in] srcALen length of the first input sequence |
| @param[in] pSrcB points to the second input sequence |
| @param[in] srcBLen length of the second input sequence |
| @param[out] pDst points to the location where the output result is written. Length 2 * max(srcALen, srcBLen) - 1. |
| @param[in] pScratch points to scratch buffer of size max(srcALen, srcBLen) + 2*min(srcALen, srcBLen) - 2. |
| @return none |
| |
| @par Scaling and Overflow Behavior |
| The function is implemented using a 64-bit internal accumulator. |
| Both inputs are in 1.15 format and multiplications yield a 2.30 result. |
| The 2.30 intermediate results are accumulated in a 64-bit accumulator in 34.30 format. |
| This approach provides 33 guard bits and there is no risk of overflow. |
| The 34.30 result is then truncated to 34.15 format by discarding the low 15 bits and then saturated to 1.15 format. |
| |
| @remark |
| Refer to \ref arm_correlate_fast_q15() for a faster but less precise version of this function. |
| */ |
| |
| void arm_correlate_opt_q15( |
| const q15_t * pSrcA, |
| uint32_t srcALen, |
| const q15_t * pSrcB, |
| uint32_t srcBLen, |
| q15_t * pDst, |
| q15_t * pScratch) |
| { |
| q63_t acc0; /* Accumulators */ |
| q15_t *pOut = pDst; /* Output pointer */ |
| q15_t *pScr1; /* Temporary pointer for scratch1 */ |
| const q15_t *pIn1; /* InputA pointer */ |
| const q15_t *pIn2; /* InputB pointer */ |
| const q15_t *py; /* Intermediate inputB pointer */ |
| uint32_t j, blkCnt, outBlockSize; /* Loop counter */ |
| int32_t inc = 1; /* Output pointer increment */ |
| uint32_t tapCnt; |
| |
| #if defined (ARM_MATH_LOOPUNROLL) |
| q63_t acc1, acc2, acc3; /* Accumulators */ |
| q31_t x1, x2, x3; /* Temporary variables for holding input1 and input2 values */ |
| q31_t y1, y2; /* State variables */ |
| #endif |
| |
| /* 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 the destination pointer modifier, inc is set to -1 */ |
| /* If srcALen > srcBLen, zero pad has to be done to srcB 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 */ |
| if (srcALen >= srcBLen) |
| { |
| /* Initialization of inputA pointer */ |
| pIn1 = pSrcA; |
| |
| /* Initialization of inputB pointer */ |
| pIn2 = pSrcB; |
| |
| /* Number of output samples is calculated */ |
| outBlockSize = (srcALen * 2U) - 1U; |
| |
| /* When srcALen > srcBLen, zero padding is done to srcB |
| * to make their lengths equal. |
| * Instead, (outBlockSize - (srcALen + srcBLen - 1)) |
| * number of output samples are made zero */ |
| j = outBlockSize - (srcALen + (srcBLen - 1U)); |
| |
| /* Updating the pointer position to non zero value */ |
| pOut += j; |
| } |
| else |
| { |
| /* Initialization of inputA pointer */ |
| pIn1 = pSrcB; |
| |
| /* Initialization of inputB pointer */ |
| pIn2 = pSrcA; |
| |
| /* srcBLen is always considered as shorter or equal to srcALen */ |
| j = srcBLen; |
| srcBLen = srcALen; |
| srcALen = j; |
| |
| /* CORR(x, y) = Reverse order(CORR(y, x)) */ |
| /* Hence set the destination pointer to point to the last output sample */ |
| pOut = pDst + ((srcALen + srcBLen) - 2U); |
| |
| /* Destination address modifier is set to -1 */ |
| inc = -1; |
| } |
| |
| pScr1 = pScratch; |
| |
| /* Fill (srcBLen - 1U) zeros in scratch buffer */ |
| arm_fill_q15(0, pScr1, (srcBLen - 1U)); |
| |
| /* Update temporary scratch pointer */ |
| pScr1 += (srcBLen - 1U); |
| |
| /* Copy (srcALen) samples in scratch buffer */ |
| arm_copy_q15(pIn1, pScr1, srcALen); |
| |
| /* Update pointers */ |
| pScr1 += srcALen; |
| |
| |
| /* Fill (srcBLen - 1U) zeros at end of scratch buffer */ |
| arm_fill_q15(0, pScr1, (srcBLen - 1U)); |
| |
| /* Update pointer */ |
| pScr1 += (srcBLen - 1U); |
| |
| /* Temporary pointer for scratch2 */ |
| py = pIn2; |
| |
| |
| /* Actual correlation process starts here */ |
| #if defined (ARM_MATH_LOOPUNROLL) |
| |
| /* Loop unrolling: Compute 4 outputs at a time */ |
| blkCnt = (srcALen + srcBLen - 1U) >> 2; |
| |
| while (blkCnt > 0) |
| { |
| /* Initialze temporary scratch pointer as scratch1 */ |
| pScr1 = pScratch; |
| |
| /* Clear Accumlators */ |
| acc0 = 0; |
| acc1 = 0; |
| acc2 = 0; |
| acc3 = 0; |
| |
| /* Read two samples from scratch1 buffer */ |
| x1 = read_q15x2_ia (&pScr1); |
| |
| /* Read next two samples from scratch1 buffer */ |
| x2 = read_q15x2_ia (&pScr1); |
| |
| tapCnt = (srcBLen) >> 2U; |
| |
| while (tapCnt > 0U) |
| { |
| /* Read four samples from smaller buffer */ |
| y1 = read_q15x2_ia ((q15_t **) &pIn2); |
| y2 = read_q15x2_ia ((q15_t **) &pIn2); |
| |
| /* multiply and accumlate */ |
| acc0 = __SMLALD(x1, y1, acc0); |
| acc2 = __SMLALD(x2, y1, acc2); |
| |
| /* pack input data */ |
| #ifndef ARM_MATH_BIG_ENDIAN |
| x3 = __PKHBT(x2, x1, 0); |
| #else |
| x3 = __PKHBT(x1, x2, 0); |
| #endif |
| |
| /* multiply and accumlate */ |
| acc1 = __SMLALDX(x3, y1, acc1); |
| |
| /* Read next two samples from scratch1 buffer */ |
| x1 = read_q15x2_ia (&pScr1); |
| |
| /* multiply and accumlate */ |
| acc0 = __SMLALD(x2, y2, acc0); |
| acc2 = __SMLALD(x1, y2, acc2); |
| |
| /* pack input data */ |
| #ifndef ARM_MATH_BIG_ENDIAN |
| x3 = __PKHBT(x1, x2, 0); |
| #else |
| x3 = __PKHBT(x2, x1, 0); |
| #endif |
| |
| acc3 = __SMLALDX(x3, y1, acc3); |
| acc1 = __SMLALDX(x3, y2, acc1); |
| |
| x2 = read_q15x2_ia (&pScr1); |
| |
| #ifndef ARM_MATH_BIG_ENDIAN |
| x3 = __PKHBT(x2, x1, 0); |
| #else |
| x3 = __PKHBT(x1, x2, 0); |
| #endif |
| |
| acc3 = __SMLALDX(x3, y2, acc3); |
| |
| /* Decrement loop counter */ |
| tapCnt--; |
| } |
| |
| /* Update scratch pointer for remaining samples of smaller length sequence */ |
| pScr1 -= 4U; |
| |
| /* apply same above for remaining samples of smaller length sequence */ |
| tapCnt = (srcBLen) & 3U; |
| |
| while (tapCnt > 0U) |
| { |
| /* accumlate the results */ |
| acc0 += (*pScr1++ * *pIn2); |
| acc1 += (*pScr1++ * *pIn2); |
| acc2 += (*pScr1++ * *pIn2); |
| acc3 += (*pScr1++ * *pIn2++); |
| |
| pScr1 -= 3U; |
| |
| /* Decrement loop counter */ |
| tapCnt--; |
| } |
| |
| blkCnt--; |
| |
| |
| /* Store the results in the accumulators in the destination buffer. */ |
| *pOut = (__SSAT(acc0 >> 15U, 16)); |
| pOut += inc; |
| *pOut = (__SSAT(acc1 >> 15U, 16)); |
| pOut += inc; |
| *pOut = (__SSAT(acc2 >> 15U, 16)); |
| pOut += inc; |
| *pOut = (__SSAT(acc3 >> 15U, 16)); |
| pOut += inc; |
| |
| /* Initialization of inputB pointer */ |
| pIn2 = py; |
| |
| pScratch += 4U; |
| } |
| |
| |
| /* Loop unrolling: Compute remaining outputs */ |
| blkCnt = (srcALen + srcBLen - 1U) & 0x3; |
| |
| #else |
| |
| /* Initialize blkCnt with number of samples */ |
| blkCnt = (srcALen + srcBLen - 1U); |
| |
| #endif /* #if defined (ARM_MATH_LOOPUNROLL) */ |
| |
| /* Calculate correlation for remaining samples of Bigger length sequence */ |
| while (blkCnt > 0) |
| { |
| /* Initialze temporary scratch pointer as scratch1 */ |
| pScr1 = pScratch; |
| |
| /* Clear Accumlators */ |
| acc0 = 0; |
| |
| tapCnt = (srcBLen) >> 1U; |
| |
| while (tapCnt > 0U) |
| { |
| |
| /* Read next two samples from scratch1 buffer */ |
| acc0 += (*pScr1++ * *pIn2++); |
| acc0 += (*pScr1++ * *pIn2++); |
| |
| /* Decrement loop counter */ |
| tapCnt--; |
| } |
| |
| tapCnt = (srcBLen) & 1U; |
| |
| /* apply same above for remaining samples of smaller length sequence */ |
| while (tapCnt > 0U) |
| { |
| /* accumlate the results */ |
| acc0 += (*pScr1++ * *pIn2++); |
| |
| /* Decrement loop counter */ |
| tapCnt--; |
| } |
| |
| blkCnt--; |
| |
| /* The result is in 2.30 format. Convert to 1.15 with saturation. |
| Then store the output in the destination buffer. */ |
| *pOut = (q15_t) (__SSAT((acc0 >> 15), 16)); |
| pOut += inc; |
| |
| /* Initialization of inputB pointer */ |
| pIn2 = py; |
| |
| pScratch += 1U; |
| } |
| |
| } |
| |
| /** |
| @} end of Corr group |
| */ |