123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122112311241125112611271128112911301131113211331134113511361137113811391140114111421143114411451146114711481149115011511152115311541155115611571158115911601161116211631164116511661167116811691170117111721173117411751176117711781179118011811182118311841185118611871188118911901191119211931194119511961197119811991200120112021203120412051206120712081209121012111212121312141215121612171218121912201221122212231224122512261227122812291230123112321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272127312741275127612771278127912801281128212831284128512861287128812891290129112921293129412951296129712981299130013011302130313041305130613071308130913101311131213131314131513161317131813191320132113221323132413251326132713281329133013311332133313341335133613371338133913401341134213431344134513461347134813491350135113521353135413551356135713581359136013611362136313641365136613671368136913701371137213731374137513761377137813791380138113821383138413851386138713881389139013911392139313941395139613971398139914001401140214031404140514061407140814091410 |
- /* ----------------------------------------------------------------------
- * Copyright (C) 2010-2013 ARM Limited. All rights reserved.
- *
- * $Date: 17. January 2013
- * $Revision: V1.4.1
- *
- * Project: CMSIS DSP Library
- * Title: arm_conv_fast_q15.c
- *
- * Description: Fast Q15 Convolution.
- *
- * Target Processor: Cortex-M4/Cortex-M3
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted provided that the following conditions
- * are met:
- * - Redistributions of source code must retain the above copyright
- * notice, this list of conditions and the following disclaimer.
- * - Redistributions in binary form must reproduce the above copyright
- * notice, this list of conditions and the following disclaimer in
- * the documentation and/or other materials provided with the
- * distribution.
- * - Neither the name of ARM LIMITED nor the names of its contributors
- * may be used to endorse or promote products derived from this
- * software without specific prior written permission.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
- * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
- * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
- * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
- * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
- * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
- * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
- * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
- * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
- * POSSIBILITY OF SUCH DAMAGE.
- * -------------------------------------------------------------------- */
-
- #include "arm_math.h"
-
- /**
- * @ingroup groupFilters
- */
-
- /**
- * @addtogroup Conv
- * @{
- */
-
- /**
- * @brief Convolution of Q15 sequences (fast version) for Cortex-M3 and Cortex-M4.
- * @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 srcALen+srcBLen-1.
- * @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. There is no saturation on intermediate additions.
- * Thus, if the accumulator overflows it wraps around and distorts the result.
- * The input signals should be scaled down to avoid intermediate overflows.
- * Scale down the inputs by log2(min(srcALen, srcBLen)) (log2 is read as log to the base 2) times to avoid overflows,
- * as maximum of min(srcALen, srcBLen) number of additions are carried internally.
- * The 2.30 accumulator is right shifted by 15 bits and then saturated to 1.15 format to yield the final result.
- *
- * \par
- * See <code>arm_conv_q15()</code> for a slower implementation of this function which uses 64-bit accumulation to avoid wrap around distortion.
- */
-
- void arm_conv_fast_q15(
- q15_t * pSrcA,
- uint32_t srcALen,
- q15_t * pSrcB,
- uint32_t srcBLen,
- q15_t * pDst)
- {
- #ifndef UNALIGNED_SUPPORT_DISABLE
- q15_t *pIn1; /* inputA pointer */
- q15_t *pIn2; /* inputB pointer */
- q15_t *pOut = pDst; /* output pointer */
- q31_t sum, acc0, acc1, acc2, acc3; /* Accumulator */
- q15_t *px; /* Intermediate inputA pointer */
- q15_t *py; /* Intermediate inputB pointer */
- q15_t *pSrc1, *pSrc2; /* Intermediate pointers */
- q31_t x0, x1, x2, x3, c0; /* Temporary variables to hold state and coefficient values */
- uint32_t blockSize1, blockSize2, blockSize3, j, k, count, blkCnt; /* loop counter */
-
- /* 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 */
- if(srcALen >= srcBLen)
- {
- /* Initialization of inputA pointer */
- pIn1 = pSrcA;
-
- /* Initialization of inputB pointer */
- pIn2 = pSrcB;
- }
- 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;
- }
-
- /* conv(x,y) at n = x[n] * y[0] + x[n-1] * y[1] + x[n-2] * y[2] + ...+ x[n-N+1] * y[N -1] */
- /* The function is internally
- * divided into three stages according to the number of multiplications that has to be
- * taken place between inputA samples and inputB samples. In the first stage of the
- * algorithm, the multiplications increase by one for every iteration.
- * In the second stage of the algorithm, srcBLen number of multiplications are done.
- * In the third stage of the algorithm, the multiplications decrease by one
- * for every iteration. */
-
- /* The algorithm is implemented in three stages.
- The loop counters of each stage is initiated here. */
- blockSize1 = srcBLen - 1u;
- blockSize2 = srcALen - (srcBLen - 1u);
- blockSize3 = blockSize1;
-
- /* --------------------------
- * Initializations of stage1
- * -------------------------*/
-
- /* sum = x[0] * y[0]
- * sum = x[0] * y[1] + x[1] * y[0]
- * ....
- * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0]
- */
-
- /* In this stage the MAC operations are increased by 1 for every iteration.
- The count variable holds the number of MAC operations performed */
- count = 1u;
-
- /* Working pointer of inputA */
- px = pIn1;
-
- /* Working pointer of inputB */
- py = pIn2;
-
-
- /* ------------------------
- * Stage1 process
- * ----------------------*/
-
- /* For loop unrolling by 4, this stage is divided into two. */
- /* First part of this stage computes the MAC operations less than 4 */
- /* Second part of this stage computes the MAC operations greater than or equal to 4 */
-
- /* The first part of the stage starts here */
- while((count < 4u) && (blockSize1 > 0u))
- {
- /* Accumulator is made zero for every iteration */
- sum = 0;
-
- /* Loop over number of MAC operations between
- * inputA samples and inputB samples */
- k = count;
-
- while(k > 0u)
- {
- /* Perform the multiply-accumulates */
- sum = __SMLAD(*px++, *py--, sum);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* Store the result in the accumulator in the destination buffer. */
- *pOut++ = (q15_t) (sum >> 15);
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- py = pIn2 + count;
- px = pIn1;
-
- /* Increment the MAC count */
- count++;
-
- /* Decrement the loop counter */
- blockSize1--;
- }
-
- /* The second part of the stage starts here */
- /* The internal loop, over count, is unrolled by 4 */
- /* To, read the last two inputB samples using SIMD:
- * y[srcBLen] and y[srcBLen-1] coefficients, py is decremented by 1 */
- py = py - 1;
-
- while(blockSize1 > 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 */
- /* x[0], x[1] are multiplied with y[srcBLen - 1], y[srcBLen - 2] respectively */
- sum = __SMLADX(*__SIMD32(px)++, *__SIMD32(py)--, sum);
- /* x[2], x[3] are multiplied with y[srcBLen - 3], y[srcBLen - 4] respectively */
- sum = __SMLADX(*__SIMD32(px)++, *__SIMD32(py)--, sum);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* For the next MAC operations, the pointer py is used without SIMD
- * So, py is incremented by 1 */
- py = py + 1u;
-
- /* 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 = __SMLAD(*px++, *py--, sum);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* Store the result in the accumulator in the destination buffer. */
- *pOut++ = (q15_t) (sum >> 15);
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- py = pIn2 + (count - 1u);
- px = pIn1;
-
- /* Increment the MAC count */
- count++;
-
- /* Decrement the loop counter */
- blockSize1--;
- }
-
- /* --------------------------
- * Initializations of stage2
- * ------------------------*/
-
- /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0]
- * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0]
- * ....
- * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0]
- */
-
- /* Working pointer of inputA */
- px = pIn1;
-
- /* Working pointer of inputB */
- pSrc2 = pIn2 + (srcBLen - 1u);
- py = pSrc2;
-
- /* count is the index by which the pointer pIn1 to be incremented */
- count = 0u;
-
-
- /* --------------------
- * Stage2 process
- * -------------------*/
-
- /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
- * So, to loop unroll over blockSize2,
- * srcBLen should be greater than or equal to 4 */
- if(srcBLen >= 4u)
- {
- /* Loop unroll over blockSize2, by 4 */
- blkCnt = blockSize2 >> 2u;
-
- while(blkCnt > 0u)
- {
- py = py - 1u;
-
- /* Set all accumulators to zero */
- acc0 = 0;
- acc1 = 0;
- acc2 = 0;
- acc3 = 0;
-
-
- /* read x[0], x[1] samples */
- x0 = *__SIMD32(px);
- /* read x[1], x[2] samples */
- x1 = _SIMD32_OFFSET(px+1);
- px+= 2u;
-
-
- /* 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. */
- do
- {
- /* Read the last two inputB samples using SIMD:
- * y[srcBLen - 1] and y[srcBLen - 2] */
- c0 = *__SIMD32(py)--;
-
- /* acc0 += x[0] * y[srcBLen - 1] + x[1] * y[srcBLen - 2] */
- acc0 = __SMLADX(x0, c0, acc0);
-
- /* acc1 += x[1] * y[srcBLen - 1] + x[2] * y[srcBLen - 2] */
- acc1 = __SMLADX(x1, c0, acc1);
-
- /* Read x[2], x[3] */
- x2 = *__SIMD32(px);
-
- /* Read x[3], x[4] */
- x3 = _SIMD32_OFFSET(px+1);
-
- /* acc2 += x[2] * y[srcBLen - 1] + x[3] * y[srcBLen - 2] */
- acc2 = __SMLADX(x2, c0, acc2);
-
- /* acc3 += x[3] * y[srcBLen - 1] + x[4] * y[srcBLen - 2] */
- acc3 = __SMLADX(x3, c0, acc3);
-
- /* Read y[srcBLen - 3] and y[srcBLen - 4] */
- c0 = *__SIMD32(py)--;
-
- /* acc0 += x[2] * y[srcBLen - 3] + x[3] * y[srcBLen - 4] */
- acc0 = __SMLADX(x2, c0, acc0);
-
- /* acc1 += x[3] * y[srcBLen - 3] + x[4] * y[srcBLen - 4] */
- acc1 = __SMLADX(x3, c0, acc1);
-
- /* Read x[4], x[5] */
- x0 = _SIMD32_OFFSET(px+2);
-
- /* Read x[5], x[6] */
- x1 = _SIMD32_OFFSET(px+3);
- px += 4u;
-
- /* acc2 += x[4] * y[srcBLen - 3] + x[5] * y[srcBLen - 4] */
- acc2 = __SMLADX(x0, c0, acc2);
-
- /* acc3 += x[5] * y[srcBLen - 3] + x[6] * y[srcBLen - 4] */
- acc3 = __SMLADX(x1, c0, acc3);
-
- } while(--k);
-
- /* For the next MAC operations, SIMD is not used
- * So, the 16 bit pointer if inputB, py is updated */
-
- /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
- ** No loop unrolling is used. */
- k = srcBLen % 0x4u;
-
- if(k == 1u)
- {
- /* Read y[srcBLen - 5] */
- c0 = *(py+1);
-
- #ifdef ARM_MATH_BIG_ENDIAN
-
- c0 = c0 << 16u;
-
- #else
-
- c0 = c0 & 0x0000FFFF;
-
- #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
-
- /* Read x[7] */
- x3 = *__SIMD32(px);
- px++;
-
- /* Perform the multiply-accumulates */
- acc0 = __SMLAD(x0, c0, acc0);
- acc1 = __SMLAD(x1, c0, acc1);
- acc2 = __SMLADX(x1, c0, acc2);
- acc3 = __SMLADX(x3, c0, acc3);
- }
-
- if(k == 2u)
- {
- /* Read y[srcBLen - 5], y[srcBLen - 6] */
- c0 = _SIMD32_OFFSET(py);
-
- /* Read x[7], x[8] */
- x3 = *__SIMD32(px);
-
- /* Read x[9] */
- x2 = _SIMD32_OFFSET(px+1);
- px += 2u;
-
- /* Perform the multiply-accumulates */
- acc0 = __SMLADX(x0, c0, acc0);
- acc1 = __SMLADX(x1, c0, acc1);
- acc2 = __SMLADX(x3, c0, acc2);
- acc3 = __SMLADX(x2, c0, acc3);
- }
-
- if(k == 3u)
- {
- /* Read y[srcBLen - 5], y[srcBLen - 6] */
- c0 = _SIMD32_OFFSET(py);
-
- /* Read x[7], x[8] */
- x3 = *__SIMD32(px);
-
- /* Read x[9] */
- x2 = _SIMD32_OFFSET(px+1);
-
- /* Perform the multiply-accumulates */
- acc0 = __SMLADX(x0, c0, acc0);
- acc1 = __SMLADX(x1, c0, acc1);
- acc2 = __SMLADX(x3, c0, acc2);
- acc3 = __SMLADX(x2, c0, acc3);
-
- /* Read y[srcBLen - 7] */
- c0 = *(py-1);
- #ifdef ARM_MATH_BIG_ENDIAN
-
- c0 = c0 << 16u;
- #else
-
- c0 = c0 & 0x0000FFFF;
- #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
-
- /* Read x[10] */
- x3 = _SIMD32_OFFSET(px+2);
- px += 3u;
-
- /* Perform the multiply-accumulates */
- acc0 = __SMLADX(x1, c0, acc0);
- acc1 = __SMLAD(x2, c0, acc1);
- acc2 = __SMLADX(x2, c0, acc2);
- acc3 = __SMLADX(x3, c0, acc3);
- }
-
- /* Store the results in the accumulators in the destination buffer. */
- #ifndef ARM_MATH_BIG_ENDIAN
-
- *__SIMD32(pOut)++ = __PKHBT((acc0 >> 15), (acc1 >> 15), 16);
- *__SIMD32(pOut)++ = __PKHBT((acc2 >> 15), (acc3 >> 15), 16);
-
- #else
-
- *__SIMD32(pOut)++ = __PKHBT((acc1 >> 15), (acc0 >> 15), 16);
- *__SIMD32(pOut)++ = __PKHBT((acc3 >> 15), (acc2 >> 15), 16);
-
- #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
-
- /* Increment the pointer pIn1 index, count by 4 */
- count += 4u;
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- px = pIn1 + count;
- py = pSrc2;
-
- /* 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 += ((q31_t) * px++ * *py--);
- sum += ((q31_t) * px++ * *py--);
- sum += ((q31_t) * px++ * *py--);
- sum += ((q31_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 += ((q31_t) * px++ * *py--);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* Store the result in the accumulator in the destination buffer. */
- *pOut++ = (q15_t) (sum >> 15);
-
- /* Increment the pointer pIn1 index, count by 1 */
- count++;
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- px = pIn1 + count;
- py = pSrc2;
-
- /* 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;
-
- /* srcBLen number of MACS should be performed */
- k = srcBLen;
-
- while(k > 0u)
- {
- /* Perform the multiply-accumulate */
- sum += ((q31_t) * px++ * *py--);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* Store the result in the accumulator in the destination buffer. */
- *pOut++ = (q15_t) (sum >> 15);
-
- /* Increment the MAC count */
- count++;
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- px = pIn1 + count;
- py = pSrc2;
-
- /* Decrement the loop counter */
- blkCnt--;
- }
- }
-
-
- /* --------------------------
- * Initializations of stage3
- * -------------------------*/
-
- /* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1]
- * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2]
- * ....
- * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2]
- * sum += x[srcALen-1] * y[srcBLen-1]
- */
-
- /* In this stage the MAC operations are decreased by 1 for every iteration.
- The blockSize3 variable holds the number of MAC operations performed */
-
- /* Working pointer of inputA */
- pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
- px = pSrc1;
-
- /* Working pointer of inputB */
- pSrc2 = pIn2 + (srcBLen - 1u);
- pIn2 = pSrc2 - 1u;
- py = pIn2;
-
- /* -------------------
- * Stage3 process
- * ------------------*/
-
- /* For loop unrolling by 4, this stage is divided into two. */
- /* First part of this stage computes the MAC operations greater than 4 */
- /* Second part of this stage computes the MAC operations less than or equal to 4 */
-
- /* The first part of the stage starts here */
- j = blockSize3 >> 2u;
-
- while((j > 0u) && (blockSize3 > 0u))
- {
- /* Accumulator is made zero for every iteration */
- sum = 0;
-
- /* Apply loop unrolling and compute 4 MACs simultaneously. */
- k = blockSize3 >> 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)
- {
- /* x[srcALen - srcBLen + 1], x[srcALen - srcBLen + 2] are multiplied
- * with y[srcBLen - 1], y[srcBLen - 2] respectively */
- sum = __SMLADX(*__SIMD32(px)++, *__SIMD32(py)--, sum);
- /* x[srcALen - srcBLen + 3], x[srcALen - srcBLen + 4] are multiplied
- * with y[srcBLen - 3], y[srcBLen - 4] respectively */
- sum = __SMLADX(*__SIMD32(px)++, *__SIMD32(py)--, sum);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* For the next MAC operations, the pointer py is used without SIMD
- * So, py is incremented by 1 */
- py = py + 1u;
-
- /* If the blockSize3 is not a multiple of 4, compute any remaining MACs here.
- ** No loop unrolling is used. */
- k = blockSize3 % 0x4u;
-
- while(k > 0u)
- {
- /* sum += x[srcALen - srcBLen + 5] * y[srcBLen - 5] */
- sum = __SMLAD(*px++, *py--, sum);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* Store the result in the accumulator in the destination buffer. */
- *pOut++ = (q15_t) (sum >> 15);
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- px = ++pSrc1;
- py = pIn2;
-
- /* Decrement the loop counter */
- blockSize3--;
-
- j--;
- }
-
- /* The second part of the stage starts here */
- /* SIMD is not used for the next MAC operations,
- * so pointer py is updated to read only one sample at a time */
- py = py + 1u;
-
- while(blockSize3 > 0u)
- {
- /* Accumulator is made zero for every iteration */
- sum = 0;
-
- /* Apply loop unrolling and compute 4 MACs simultaneously. */
- k = blockSize3;
-
- while(k > 0u)
- {
- /* Perform the multiply-accumulates */
- /* sum += x[srcALen-1] * y[srcBLen-1] */
- sum = __SMLAD(*px++, *py--, sum);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* Store the result in the accumulator in the destination buffer. */
- *pOut++ = (q15_t) (sum >> 15);
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- px = ++pSrc1;
- py = pSrc2;
-
- /* Decrement the loop counter */
- blockSize3--;
- }
-
- #else
- q15_t *pIn1; /* inputA pointer */
- q15_t *pIn2; /* inputB pointer */
- q15_t *pOut = pDst; /* output pointer */
- q31_t sum, acc0, acc1, acc2, acc3; /* Accumulator */
- q15_t *px; /* Intermediate inputA pointer */
- q15_t *py; /* Intermediate inputB pointer */
- q15_t *pSrc1, *pSrc2; /* Intermediate pointers */
- q31_t x0, x1, x2, x3, c0; /* Temporary variables to hold state and coefficient values */
- uint32_t blockSize1, blockSize2, blockSize3, j, k, count, blkCnt; /* loop counter */
- q15_t a, b;
-
- /* 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 */
- if(srcALen >= srcBLen)
- {
- /* Initialization of inputA pointer */
- pIn1 = pSrcA;
-
- /* Initialization of inputB pointer */
- pIn2 = pSrcB;
- }
- 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;
- }
-
- /* conv(x,y) at n = x[n] * y[0] + x[n-1] * y[1] + x[n-2] * y[2] + ...+ x[n-N+1] * y[N -1] */
- /* The function is internally
- * divided into three stages according to the number of multiplications that has to be
- * taken place between inputA samples and inputB samples. In the first stage of the
- * algorithm, the multiplications increase by one for every iteration.
- * In the second stage of the algorithm, srcBLen number of multiplications are done.
- * In the third stage of the algorithm, the multiplications decrease by one
- * for every iteration. */
-
- /* The algorithm is implemented in three stages.
- The loop counters of each stage is initiated here. */
- blockSize1 = srcBLen - 1u;
- blockSize2 = srcALen - (srcBLen - 1u);
- blockSize3 = blockSize1;
-
- /* --------------------------
- * Initializations of stage1
- * -------------------------*/
-
- /* sum = x[0] * y[0]
- * sum = x[0] * y[1] + x[1] * y[0]
- * ....
- * sum = x[0] * y[srcBlen - 1] + x[1] * y[srcBlen - 2] +...+ x[srcBLen - 1] * y[0]
- */
-
- /* In this stage the MAC operations are increased by 1 for every iteration.
- The count variable holds the number of MAC operations performed */
- count = 1u;
-
- /* Working pointer of inputA */
- px = pIn1;
-
- /* Working pointer of inputB */
- py = pIn2;
-
-
- /* ------------------------
- * Stage1 process
- * ----------------------*/
-
- /* For loop unrolling by 4, this stage is divided into two. */
- /* First part of this stage computes the MAC operations less than 4 */
- /* Second part of this stage computes the MAC operations greater than or equal to 4 */
-
- /* The first part of the stage starts here */
- while((count < 4u) && (blockSize1 > 0u))
- {
- /* Accumulator is made zero for every iteration */
- sum = 0;
-
- /* Loop over number of MAC operations between
- * inputA samples and inputB samples */
- k = count;
-
- while(k > 0u)
- {
- /* Perform the multiply-accumulates */
- sum += ((q31_t) * px++ * *py--);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* Store the result in the accumulator in the destination buffer. */
- *pOut++ = (q15_t) (sum >> 15);
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- py = pIn2 + count;
- px = pIn1;
-
- /* Increment the MAC count */
- count++;
-
- /* Decrement the loop counter */
- blockSize1--;
- }
-
- /* The second part of the stage starts here */
- /* The internal loop, over count, is unrolled by 4 */
- /* To, read the last two inputB samples using SIMD:
- * y[srcBLen] and y[srcBLen-1] coefficients, py is decremented by 1 */
- py = py - 1;
-
- while(blockSize1 > 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. */
- py++;
-
- while(k > 0u)
- {
- /* Perform the multiply-accumulates */
- sum += ((q31_t) * px++ * *py--);
- sum += ((q31_t) * px++ * *py--);
- sum += ((q31_t) * px++ * *py--);
- sum += ((q31_t) * px++ * *py--);
-
- /* 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 += ((q31_t) * px++ * *py--);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* Store the result in the accumulator in the destination buffer. */
- *pOut++ = (q15_t) (sum >> 15);
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- py = pIn2 + (count - 1u);
- px = pIn1;
-
- /* Increment the MAC count */
- count++;
-
- /* Decrement the loop counter */
- blockSize1--;
- }
-
- /* --------------------------
- * Initializations of stage2
- * ------------------------*/
-
- /* sum = x[0] * y[srcBLen-1] + x[1] * y[srcBLen-2] +...+ x[srcBLen-1] * y[0]
- * sum = x[1] * y[srcBLen-1] + x[2] * y[srcBLen-2] +...+ x[srcBLen] * y[0]
- * ....
- * sum = x[srcALen-srcBLen-2] * y[srcBLen-1] + x[srcALen] * y[srcBLen-2] +...+ x[srcALen-1] * y[0]
- */
-
- /* Working pointer of inputA */
- px = pIn1;
-
- /* Working pointer of inputB */
- pSrc2 = pIn2 + (srcBLen - 1u);
- py = pSrc2;
-
- /* count is the index by which the pointer pIn1 to be incremented */
- count = 0u;
-
-
- /* --------------------
- * Stage2 process
- * -------------------*/
-
- /* Stage2 depends on srcBLen as in this stage srcBLen number of MACS are performed.
- * So, to loop unroll over blockSize2,
- * srcBLen should be greater than or equal to 4 */
- if(srcBLen >= 4u)
- {
- /* Loop unroll over blockSize2, by 4 */
- blkCnt = blockSize2 >> 2u;
-
- while(blkCnt > 0u)
- {
- py = py - 1u;
-
- /* Set all accumulators to zero */
- acc0 = 0;
- acc1 = 0;
- acc2 = 0;
- acc3 = 0;
-
- /* read x[0], x[1] samples */
- a = *px++;
- b = *px++;
-
- #ifndef ARM_MATH_BIG_ENDIAN
-
- x0 = __PKHBT(a, b, 16);
- a = *px;
- x1 = __PKHBT(b, a, 16);
-
- #else
-
- x0 = __PKHBT(b, a, 16);
- a = *px;
- x1 = __PKHBT(a, b, 16);
-
- #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
-
- /* 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. */
- do
- {
- /* Read the last two inputB samples using SIMD:
- * y[srcBLen - 1] and y[srcBLen - 2] */
- a = *py;
- b = *(py+1);
- py -= 2;
-
- #ifndef ARM_MATH_BIG_ENDIAN
-
- c0 = __PKHBT(a, b, 16);
-
- #else
-
- c0 = __PKHBT(b, a, 16);;
-
- #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
-
- /* acc0 += x[0] * y[srcBLen - 1] + x[1] * y[srcBLen - 2] */
- acc0 = __SMLADX(x0, c0, acc0);
-
- /* acc1 += x[1] * y[srcBLen - 1] + x[2] * y[srcBLen - 2] */
- acc1 = __SMLADX(x1, c0, acc1);
-
- a = *px;
- b = *(px + 1);
-
- #ifndef ARM_MATH_BIG_ENDIAN
-
- x2 = __PKHBT(a, b, 16);
- a = *(px + 2);
- x3 = __PKHBT(b, a, 16);
-
- #else
-
- x2 = __PKHBT(b, a, 16);
- a = *(px + 2);
- x3 = __PKHBT(a, b, 16);
-
- #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
-
- /* acc2 += x[2] * y[srcBLen - 1] + x[3] * y[srcBLen - 2] */
- acc2 = __SMLADX(x2, c0, acc2);
-
- /* acc3 += x[3] * y[srcBLen - 1] + x[4] * y[srcBLen - 2] */
- acc3 = __SMLADX(x3, c0, acc3);
-
- /* Read y[srcBLen - 3] and y[srcBLen - 4] */
- a = *py;
- b = *(py+1);
- py -= 2;
-
- #ifndef ARM_MATH_BIG_ENDIAN
-
- c0 = __PKHBT(a, b, 16);
-
- #else
-
- c0 = __PKHBT(b, a, 16);;
-
- #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
-
- /* acc0 += x[2] * y[srcBLen - 3] + x[3] * y[srcBLen - 4] */
- acc0 = __SMLADX(x2, c0, acc0);
-
- /* acc1 += x[3] * y[srcBLen - 3] + x[4] * y[srcBLen - 4] */
- acc1 = __SMLADX(x3, c0, acc1);
-
- /* Read x[4], x[5], x[6] */
- a = *(px + 2);
- b = *(px + 3);
-
- #ifndef ARM_MATH_BIG_ENDIAN
-
- x0 = __PKHBT(a, b, 16);
- a = *(px + 4);
- x1 = __PKHBT(b, a, 16);
-
- #else
-
- x0 = __PKHBT(b, a, 16);
- a = *(px + 4);
- x1 = __PKHBT(a, b, 16);
-
- #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
-
- px += 4u;
-
- /* acc2 += x[4] * y[srcBLen - 3] + x[5] * y[srcBLen - 4] */
- acc2 = __SMLADX(x0, c0, acc2);
-
- /* acc3 += x[5] * y[srcBLen - 3] + x[6] * y[srcBLen - 4] */
- acc3 = __SMLADX(x1, c0, acc3);
-
- } while(--k);
-
- /* For the next MAC operations, SIMD is not used
- * So, the 16 bit pointer if inputB, py is updated */
-
- /* If the srcBLen is not a multiple of 4, compute any remaining MACs here.
- ** No loop unrolling is used. */
- k = srcBLen % 0x4u;
-
- if(k == 1u)
- {
- /* Read y[srcBLen - 5] */
- c0 = *(py+1);
-
- #ifdef ARM_MATH_BIG_ENDIAN
-
- c0 = c0 << 16u;
-
- #else
-
- c0 = c0 & 0x0000FFFF;
-
- #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
-
- /* Read x[7] */
- a = *px;
- b = *(px+1);
- px++;
-
- #ifndef ARM_MATH_BIG_ENDIAN
-
- x3 = __PKHBT(a, b, 16);
-
- #else
-
- x3 = __PKHBT(b, a, 16);;
-
- #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
-
-
- /* Perform the multiply-accumulates */
- acc0 = __SMLAD(x0, c0, acc0);
- acc1 = __SMLAD(x1, c0, acc1);
- acc2 = __SMLADX(x1, c0, acc2);
- acc3 = __SMLADX(x3, c0, acc3);
- }
-
- if(k == 2u)
- {
- /* Read y[srcBLen - 5], y[srcBLen - 6] */
- a = *py;
- b = *(py+1);
-
- #ifndef ARM_MATH_BIG_ENDIAN
-
- c0 = __PKHBT(a, b, 16);
-
- #else
-
- c0 = __PKHBT(b, a, 16);;
-
- #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
-
- /* Read x[7], x[8], x[9] */
- a = *px;
- b = *(px + 1);
-
- #ifndef ARM_MATH_BIG_ENDIAN
-
- x3 = __PKHBT(a, b, 16);
- a = *(px + 2);
- x2 = __PKHBT(b, a, 16);
-
- #else
-
- x3 = __PKHBT(b, a, 16);
- a = *(px + 2);
- x2 = __PKHBT(a, b, 16);
-
- #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
- px += 2u;
-
- /* Perform the multiply-accumulates */
- acc0 = __SMLADX(x0, c0, acc0);
- acc1 = __SMLADX(x1, c0, acc1);
- acc2 = __SMLADX(x3, c0, acc2);
- acc3 = __SMLADX(x2, c0, acc3);
- }
-
- if(k == 3u)
- {
- /* Read y[srcBLen - 5], y[srcBLen - 6] */
- a = *py;
- b = *(py+1);
-
- #ifndef ARM_MATH_BIG_ENDIAN
-
- c0 = __PKHBT(a, b, 16);
-
- #else
-
- c0 = __PKHBT(b, a, 16);;
-
- #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
-
- /* Read x[7], x[8], x[9] */
- a = *px;
- b = *(px + 1);
-
- #ifndef ARM_MATH_BIG_ENDIAN
-
- x3 = __PKHBT(a, b, 16);
- a = *(px + 2);
- x2 = __PKHBT(b, a, 16);
-
- #else
-
- x3 = __PKHBT(b, a, 16);
- a = *(px + 2);
- x2 = __PKHBT(a, b, 16);
-
- #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
-
- /* Perform the multiply-accumulates */
- acc0 = __SMLADX(x0, c0, acc0);
- acc1 = __SMLADX(x1, c0, acc1);
- acc2 = __SMLADX(x3, c0, acc2);
- acc3 = __SMLADX(x2, c0, acc3);
-
- /* Read y[srcBLen - 7] */
- c0 = *(py-1);
- #ifdef ARM_MATH_BIG_ENDIAN
-
- c0 = c0 << 16u;
- #else
-
- c0 = c0 & 0x0000FFFF;
- #endif /* #ifdef ARM_MATH_BIG_ENDIAN */
-
- /* Read x[10] */
- a = *(px+2);
- b = *(px+3);
-
- #ifndef ARM_MATH_BIG_ENDIAN
-
- x3 = __PKHBT(a, b, 16);
-
- #else
-
- x3 = __PKHBT(b, a, 16);;
-
- #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
-
- px += 3u;
-
- /* Perform the multiply-accumulates */
- acc0 = __SMLADX(x1, c0, acc0);
- acc1 = __SMLAD(x2, c0, acc1);
- acc2 = __SMLADX(x2, c0, acc2);
- acc3 = __SMLADX(x3, c0, acc3);
- }
-
- /* Store the results in the accumulators in the destination buffer. */
- *pOut++ = (q15_t)(acc0 >> 15);
- *pOut++ = (q15_t)(acc1 >> 15);
- *pOut++ = (q15_t)(acc2 >> 15);
- *pOut++ = (q15_t)(acc3 >> 15);
-
- /* Increment the pointer pIn1 index, count by 4 */
- count += 4u;
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- px = pIn1 + count;
- py = pSrc2;
-
- /* 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 += ((q31_t) * px++ * *py--);
- sum += ((q31_t) * px++ * *py--);
- sum += ((q31_t) * px++ * *py--);
- sum += ((q31_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 += ((q31_t) * px++ * *py--);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* Store the result in the accumulator in the destination buffer. */
- *pOut++ = (q15_t) (sum >> 15);
-
- /* Increment the pointer pIn1 index, count by 1 */
- count++;
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- px = pIn1 + count;
- py = pSrc2;
-
- /* 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;
-
- /* srcBLen number of MACS should be performed */
- k = srcBLen;
-
- while(k > 0u)
- {
- /* Perform the multiply-accumulate */
- sum += ((q31_t) * px++ * *py--);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* Store the result in the accumulator in the destination buffer. */
- *pOut++ = (q15_t) (sum >> 15);
-
- /* Increment the MAC count */
- count++;
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- px = pIn1 + count;
- py = pSrc2;
-
- /* Decrement the loop counter */
- blkCnt--;
- }
- }
-
-
- /* --------------------------
- * Initializations of stage3
- * -------------------------*/
-
- /* sum += x[srcALen-srcBLen+1] * y[srcBLen-1] + x[srcALen-srcBLen+2] * y[srcBLen-2] +...+ x[srcALen-1] * y[1]
- * sum += x[srcALen-srcBLen+2] * y[srcBLen-1] + x[srcALen-srcBLen+3] * y[srcBLen-2] +...+ x[srcALen-1] * y[2]
- * ....
- * sum += x[srcALen-2] * y[srcBLen-1] + x[srcALen-1] * y[srcBLen-2]
- * sum += x[srcALen-1] * y[srcBLen-1]
- */
-
- /* In this stage the MAC operations are decreased by 1 for every iteration.
- The blockSize3 variable holds the number of MAC operations performed */
-
- /* Working pointer of inputA */
- pSrc1 = (pIn1 + srcALen) - (srcBLen - 1u);
- px = pSrc1;
-
- /* Working pointer of inputB */
- pSrc2 = pIn2 + (srcBLen - 1u);
- pIn2 = pSrc2 - 1u;
- py = pIn2;
-
- /* -------------------
- * Stage3 process
- * ------------------*/
-
- /* For loop unrolling by 4, this stage is divided into two. */
- /* First part of this stage computes the MAC operations greater than 4 */
- /* Second part of this stage computes the MAC operations less than or equal to 4 */
-
- /* The first part of the stage starts here */
- j = blockSize3 >> 2u;
-
- while((j > 0u) && (blockSize3 > 0u))
- {
- /* Accumulator is made zero for every iteration */
- sum = 0;
-
- /* Apply loop unrolling and compute 4 MACs simultaneously. */
- k = blockSize3 >> 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. */
- py++;
-
- while(k > 0u)
- {
- sum += ((q31_t) * px++ * *py--);
- sum += ((q31_t) * px++ * *py--);
- sum += ((q31_t) * px++ * *py--);
- sum += ((q31_t) * px++ * *py--);
- /* Decrement the loop counter */
- k--;
- }
-
- /* If the blockSize3 is not a multiple of 4, compute any remaining MACs here.
- ** No loop unrolling is used. */
- k = blockSize3 % 0x4u;
-
- while(k > 0u)
- {
- /* sum += x[srcALen - srcBLen + 5] * y[srcBLen - 5] */
- sum += ((q31_t) * px++ * *py--);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* Store the result in the accumulator in the destination buffer. */
- *pOut++ = (q15_t) (sum >> 15);
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- px = ++pSrc1;
- py = pIn2;
-
- /* Decrement the loop counter */
- blockSize3--;
-
- j--;
- }
-
- /* The second part of the stage starts here */
- /* SIMD is not used for the next MAC operations,
- * so pointer py is updated to read only one sample at a time */
- py = py + 1u;
-
- while(blockSize3 > 0u)
- {
- /* Accumulator is made zero for every iteration */
- sum = 0;
-
- /* Apply loop unrolling and compute 4 MACs simultaneously. */
- k = blockSize3;
-
- while(k > 0u)
- {
- /* Perform the multiply-accumulates */
- /* sum += x[srcALen-1] * y[srcBLen-1] */
- sum += ((q31_t) * px++ * *py--);
-
- /* Decrement the loop counter */
- k--;
- }
-
- /* Store the result in the accumulator in the destination buffer. */
- *pOut++ = (q15_t) (sum >> 15);
-
- /* Update the inputA and inputB pointers for next MAC calculation */
- px = ++pSrc1;
- py = pSrc2;
-
- /* Decrement the loop counter */
- blockSize3--;
- }
-
- #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */
- }
-
- /**
- * @} end of Conv group
- */
|