arm_dot_prod_q31.c

00001 /* ----------------------------------------------------------------------
00002  * Project:      CMSIS DSP Library
00003  * Title:        arm_dot_prod_q31.c
00004  * Description:  Q31 dot product
00005  *
00006  * $Date:        27. January 2017
00007  * $Revision:    V.1.5.1
00008  *
00009  * Target Processor: Cortex-M cores
00010  * -------------------------------------------------------------------- */
00011 /*
00012  * Copyright (C) 2010-2017 ARM Limited or its affiliates. All rights reserved.
00013  *
00014  * SPDX-License-Identifier: Apache-2.0
00015  *
00016  * Licensed under the Apache License, Version 2.0 (the License); you may
00017  * not use this file except in compliance with the License.
00018  * You may obtain a copy of the License at
00019  *
00020  * www.apache.org/licenses/LICENSE-2.0
00021  *
00022  * Unless required by applicable law or agreed to in writing, software
00023  * distributed under the License is distributed on an AS IS BASIS, WITHOUT
00024  * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
00025  * See the License for the specific language governing permissions and
00026  * limitations under the License.
00027  */
00028 
00029 #include "arm_math.h"
00030 
00031 /**
00032  * @ingroup groupMath
00033  */
00034 
00035 /**
00036  * @addtogroup dot_prod
00037  * @{
00038  */
00039 
00040 /**
00041  * @brief Dot product of Q31 vectors.
00042  * @param[in]       *pSrcA points to the first input vector
00043  * @param[in]       *pSrcB points to the second input vector
00044  * @param[in]       blockSize number of samples in each vector
00045  * @param[out]      *result output result returned here
00046  * @return none.
00047  *
00048  * <b>Scaling and Overflow Behavior:</b>
00049  * \par
00050  * The intermediate multiplications are in 1.31 x 1.31 = 2.62 format and these
00051  * are truncated to 2.48 format by discarding the lower 14 bits.
00052  * The 2.48 result is then added without saturation to a 64-bit accumulator in 16.48 format.
00053  * There are 15 guard bits in the accumulator and there is no risk of overflow as long as
00054  * the length of the vectors is less than 2^16 elements.
00055  * The return result is in 16.48 format.
00056  */
00057 
00058 void arm_dot_prod_q31(
00059   q31_t * pSrcA,
00060   q31_t * pSrcB,
00061   uint32_t blockSize,
00062   q63_t * result)
00063 {
00064   q63_t sum = 0;                                 /* Temporary result storage */
00065   uint32_t blkCnt;                               /* loop counter */
00066 
00067 
00068 #if defined (ARM_MATH_DSP)
00069 
00070 /* Run the below code for Cortex-M4 and Cortex-M3 */
00071   q31_t inA1, inA2, inA3, inA4;
00072   q31_t inB1, inB2, inB3, inB4;
00073 
00074   /*loop Unrolling */
00075   blkCnt = blockSize >> 2U;
00076 
00077   /* First part of the processing with loop unrolling.  Compute 4 outputs at a time.
00078    ** a second loop below computes the remaining 1 to 3 samples. */
00079   while (blkCnt > 0U)
00080   {
00081     /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */
00082     /* Calculate dot product and then store the result in a temporary buffer. */
00083     inA1 = *pSrcA++;
00084     inA2 = *pSrcA++;
00085     inA3 = *pSrcA++;
00086     inA4 = *pSrcA++;
00087     inB1 = *pSrcB++;
00088     inB2 = *pSrcB++;
00089     inB3 = *pSrcB++;
00090     inB4 = *pSrcB++;
00091 
00092     sum += ((q63_t) inA1 * inB1) >> 14U;
00093     sum += ((q63_t) inA2 * inB2) >> 14U;
00094     sum += ((q63_t) inA3 * inB3) >> 14U;
00095     sum += ((q63_t) inA4 * inB4) >> 14U;
00096 
00097     /* Decrement the loop counter */
00098     blkCnt--;
00099   }
00100 
00101   /* If the blockSize is not a multiple of 4, compute any remaining output samples here.
00102    ** No loop unrolling is used. */
00103   blkCnt = blockSize % 0x4U;
00104 
00105 #else
00106 
00107   /* Run the below code for Cortex-M0 */
00108 
00109   /* Initialize blkCnt with number of samples */
00110   blkCnt = blockSize;
00111 
00112 #endif /* #if defined (ARM_MATH_DSP) */
00113 
00114 
00115   while (blkCnt > 0U)
00116   {
00117     /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */
00118     /* Calculate dot product and then store the result in a temporary buffer. */
00119     sum += ((q63_t) * pSrcA++ * *pSrcB++) >> 14U;
00120 
00121     /* Decrement the loop counter */
00122     blkCnt--;
00123   }
00124 
00125   /* Store the result in the destination buffer in 16.48 format */
00126   *result = sum;
00127 }
00128 
00129 /**
00130  * @} end of dot_prod group
00131  */
00132