12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697989910010110210310410510610710810911011111211311411511611711811912012112212312412512612712812913013113213313413513613713813914014114214314414514614714814915015115215315415515615715815916016116216316416516616716816917017117217317417517617717817918018118218318418518618718818919019119219319419519619719819920020120220320420520620720820921021121221321421521621721821922022122222322422522622722822923023123223323423523623723823924024124224324424524624724824925025125225325425525625725825926026126226326426526626726826927027127227327427527627727827928028128228328428528628728828929029129229329429529629729829930030130230330430530630730830931031131231331431531631731831932032132232332432532632732832933033133233333433533633733833934034134234334434534634734834935035135235335435535635735835936036136236336436536636736836937037137237337437537637737837938038138238338438538638738838939039139239339439539639739839940040140240340440540640740840941041141241341441541641741841942042142242342442542642742842943043143243343443543643743843944044144244344444544644744844945045145245345445545645745845946046146246346446546646746846947047147247347447547647747847948048148248348448548648748848949049149249349449549649749849950050150250350450550650750850951051151251351451551651751851952052152252352452552652752852953053153253353453553653753853954054154254354454554654754854955055155255355455555655755855956056156256356456556656756856957057157257357457557657757857958058158258358458558658758858959059159259359459559659759859960060160260360460560660760860961061161261361461561661761861962062162262362462562662762862963063163263363463563663763863964064164264364464564664764864965065165265365465565665765865966066166266366466566666766866967067167267367467567667767867968068168268368468568668768868969069169269369469569669769869970070170270370470570670770870971071171271371471571671771871972072172272372472572672772872973073173273373473573673773873974074174274374474574674774874975075175275375475575675775875976076176276376476576676776876977077177277377477577677777877978078178278378478578678778878979079179279379479579679779879980080180280380480580680780880981081181281381481581681781881982082182282382482582682782882983083183283383483583683783883984084184284384484584684784884985085185285385485585685785885986086186286386486586686786886987087187287387487587687787887988088188288388488588688788888989089189289389489589689789889990090190290390490590690790890991091191291391491591691791891992092192292392492592692792892993093193293393493593693793893994094194294394494594694794894995095195295395495595695795895996096196296396496596696796896997097197297397497597697797897998098198298398498598698798898999099199299399499599699799899910001001100210031004100510061007100810091010101110121013101410151016101710181019102010211022102310241025102610271028102910301031103210331034103510361037103810391040104110421043104410451046104710481049105010511052105310541055105610571058105910601061106210631064106510661067106810691070107110721073107410751076107710781079108010811082108310841085108610871088108910901091109210931094109510961097109810991100110111021103110411051106110711081109111011111112111311141115111611171118111911201121112211231124112511261127112811291130113111321133113411351136113711381139114011411142114311441145114611471148114911501151115211531154115511561157115811591160116111621163116411651166116711681169117011711172117311741175117611771178117911801181118211831184118511861187118811891190119111921193119411951196119711981199120012011202120312041205120612071208120912101211121212131214121512161217121812191220122112221223122412251226122712281229123012311232123312341235123612371238123912401241124212431244124512461247124812491250125112521253125412551256125712581259126012611262126312641265126612671268126912701271127212731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310131113121313131413151316131713181319132013211322132313241325132613271328132913301331133213331334133513361337133813391340134113421343134413451346134713481349135013511352135313541355135613571358135913601361136213631364136513661367136813691370137113721373137413751376137713781379138013811382138313841385138613871388138913901391139213931394139513961397139813991400140114021403140414051406140714081409141014111412141314141415141614171418141914201421142214231424142514261427142814291430143114321433143414351436143714381439144014411442144314441445144614471448144914501451145214531454145514561457145814591460146114621463146414651466146714681469147014711472147314741475147614771478 |
- /* ----------------------------------------------------------------------
- * Copyright (C) 2010-2013 ARM Limited. All rights reserved.
- *
- * $Date: 17. January 2013
- * $Revision: V1.4.1
- *
- * Project: CMSIS DSP Library
- * Title: arm_conv_partial_fast_q15.c
- *
- * Description: Fast Q15 Partial 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 PartialConv
- * @{
- */
-
- /**
- * @brief Partial 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.
- * @param[in] firstIndex is the first output sample to start with.
- * @param[in] numPoints is the number of output points to be computed.
- * @return Returns either ARM_MATH_SUCCESS if the function completed correctly or ARM_MATH_ARGUMENT_ERROR if the requested subset is not in the range [0 srcALen+srcBLen-2].
- *
- * See <code>arm_conv_partial_q15()</code> for a slower implementation of this function which uses a 64-bit accumulator to avoid wrap around distortion.
- */
-
-
- arm_status arm_conv_partial_fast_q15(
- q15_t * pSrcA,
- uint32_t srcALen,
- q15_t * pSrcB,
- uint32_t srcBLen,
- q15_t * pDst,
- uint32_t firstIndex,
- uint32_t numPoints)
- {
- #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;
- uint32_t j, k, count, check, blkCnt;
- int32_t blockSize1, blockSize2, blockSize3; /* loop counters */
- arm_status status; /* status of Partial convolution */
-
- /* Check for range of output samples to be calculated */
- if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u))))
- {
- /* Set status as ARM_MATH_ARGUMENT_ERROR */
- status = ARM_MATH_ARGUMENT_ERROR;
- }
- else
- {
-
- /* 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;
- }
-
- /* Conditions to check which loopCounter holds
- * the first and last indices of the output samples to be calculated. */
- check = firstIndex + numPoints;
- blockSize3 = ((int32_t) check - (int32_t) srcALen);
- blockSize3 = (blockSize3 > 0) ? blockSize3 : 0;
- blockSize1 = (((int32_t) srcBLen - 1) - (int32_t) firstIndex);
- blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1u)) ? blockSize1 :
- (int32_t) numPoints) : 0;
- blockSize2 = (int32_t) check - ((blockSize3 + blockSize1) +
- (int32_t) firstIndex);
- blockSize2 = (blockSize2 > 0) ? blockSize2 : 0;
-
- /* 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. */
-
- /* Set the output pointer to point to the firstIndex
- * of the output sample to be calculated. */
- pOut = pDst + firstIndex;
-
- /* --------------------------
- * 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.
- Since the partial convolution starts from firstIndex
- Number of Macs to be performed is firstIndex + 1 */
- count = 1u + firstIndex;
-
- /* Working pointer of inputA */
- px = pIn1;
-
- /* Working pointer of inputB */
- pSrc2 = pIn2 + firstIndex;
- py = pSrc2;
-
- /* ------------------------
- * 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 > 0))
- {
- /* 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 = ++pSrc2;
- 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 > 0)
- {
- /* 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 = ++pSrc2 - 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 = ((uint32_t) 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);
-
- 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 = (uint32_t) 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 = (uint32_t) 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 count variable holds the number of MAC operations performed */
- count = srcBLen - 1u;
-
- /* 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 = count >> 2u;
-
- while((j > 0u) && (blockSize3 > 0))
- {
- /* 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)
- {
- /* 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 count is not a multiple of 4, compute any remaining MACs here.
- ** No loop unrolling is used. */
- k = count % 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 MAC count */
- count--;
-
- /* 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 > 0)
- {
- /* Accumulator is made zero for every iteration */
- sum = 0;
-
- /* Apply loop unrolling and compute 4 MACs simultaneously. */
- k = count;
-
- 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 MAC count */
- count--;
-
- /* Decrement the loop counter */
- blockSize3--;
- }
-
- /* set status as ARM_MATH_SUCCESS */
- status = ARM_MATH_SUCCESS;
- }
-
- /* Return to application */
- return (status);
-
- #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;
- uint32_t j, k, count, check, blkCnt;
- int32_t blockSize1, blockSize2, blockSize3; /* loop counters */
- arm_status status; /* status of Partial convolution */
- q15_t a, b;
-
- /* Check for range of output samples to be calculated */
- if((firstIndex + numPoints) > ((srcALen + (srcBLen - 1u))))
- {
- /* Set status as ARM_MATH_ARGUMENT_ERROR */
- status = ARM_MATH_ARGUMENT_ERROR;
- }
- else
- {
-
- /* 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;
- }
-
- /* Conditions to check which loopCounter holds
- * the first and last indices of the output samples to be calculated. */
- check = firstIndex + numPoints;
- blockSize3 = ((int32_t) check - (int32_t) srcALen);
- blockSize3 = (blockSize3 > 0) ? blockSize3 : 0;
- blockSize1 = (((int32_t) srcBLen - 1) - (int32_t) firstIndex);
- blockSize1 = (blockSize1 > 0) ? ((check > (srcBLen - 1u)) ? blockSize1 :
- (int32_t) numPoints) : 0;
- blockSize2 = (int32_t) check - ((blockSize3 + blockSize1) +
- (int32_t) firstIndex);
- blockSize2 = (blockSize2 > 0) ? blockSize2 : 0;
-
- /* 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. */
-
- /* Set the output pointer to point to the firstIndex
- * of the output sample to be calculated. */
- pOut = pDst + firstIndex;
-
- /* --------------------------
- * 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.
- Since the partial convolution starts from firstIndex
- Number of Macs to be performed is firstIndex + 1 */
- count = 1u + firstIndex;
-
- /* Working pointer of inputA */
- px = pIn1;
-
- /* Working pointer of inputB */
- pSrc2 = pIn2 + firstIndex;
- py = pSrc2;
-
- /* ------------------------
- * 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 = ++pSrc2;
- 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 = ++pSrc2 - 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 = ((uint32_t) 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 = (uint32_t) 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 = (uint32_t) 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 count variable holds the number of MAC operations performed */
- count = srcBLen - 1u;
-
- /* 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 = count >> 2u;
-
- while((j > 0u) && (blockSize3 > 0))
- {
- /* 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 */
- px = ++pSrc1;
- py = pIn2;
-
- /* Decrement the MAC count */
- count--;
-
- /* 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 = count;
-
- 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 MAC count */
- count--;
-
- /* Decrement the loop counter */
- blockSize3--;
- }
-
- /* set status as ARM_MATH_SUCCESS */
- status = ARM_MATH_SUCCESS;
- }
-
- /* Return to application */
- return (status);
-
- #endif /* #ifndef UNALIGNED_SUPPORT_DISABLE */
- }
-
- /**
- * @} end of PartialConv group
- */
|