arm_dot_prod_q31.c

00001 /* ----------------------------------------------------------------------    
00002 * Copyright (C) 2010-2014 ARM Limited. All rights reserved.    
00003 *    
00004 * $Date:        19. March 2015
00005 * $Revision:    V.1.4.5
00006 *    
00007 * Project:      CMSIS DSP Library    
00008 * Title:        arm_dot_prod_q31.c    
00009 *    
00010 * Description:  Q31 dot product.    
00011 *    
00012 * Target Processor: Cortex-M4/Cortex-M3/Cortex-M0
00013 *  
00014 * Redistribution and use in source and binary forms, with or without 
00015 * modification, are permitted provided that the following conditions
00016 * are met:
00017 *   - Redistributions of source code must retain the above copyright
00018 *     notice, this list of conditions and the following disclaimer.
00019 *   - Redistributions in binary form must reproduce the above copyright
00020 *     notice, this list of conditions and the following disclaimer in
00021 *     the documentation and/or other materials provided with the 
00022 *     distribution.
00023 *   - Neither the name of ARM LIMITED nor the names of its contributors
00024 *     may be used to endorse or promote products derived from this
00025 *     software without specific prior written permission.
00026 *
00027 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
00028 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
00029 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
00030 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 
00031 * COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
00032 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
00033 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
00034 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
00035 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
00036 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
00037 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
00038 * POSSIBILITY OF SUCH DAMAGE. 
00039 * -------------------------------------------------------------------- */
00040 
00041 #include "arm_math.h"
00042 
00043 /**    
00044  * @ingroup groupMath    
00045  */
00046 
00047 /**    
00048  * @addtogroup dot_prod    
00049  * @{    
00050  */
00051 
00052 /**    
00053  * @brief Dot product of Q31 vectors.    
00054  * @param[in]       *pSrcA points to the first input vector    
00055  * @param[in]       *pSrcB points to the second input vector    
00056  * @param[in]       blockSize number of samples in each vector    
00057  * @param[out]      *result output result returned here    
00058  * @return none.    
00059  *    
00060  * <b>Scaling and Overflow Behavior:</b>    
00061  * \par    
00062  * The intermediate multiplications are in 1.31 x 1.31 = 2.62 format and these    
00063  * are truncated to 2.48 format by discarding the lower 14 bits.    
00064  * The 2.48 result is then added without saturation to a 64-bit accumulator in 16.48 format.    
00065  * There are 15 guard bits in the accumulator and there is no risk of overflow as long as    
00066  * the length of the vectors is less than 2^16 elements.    
00067  * The return result is in 16.48 format.    
00068  */
00069 
00070 void arm_dot_prod_q31(
00071   q31_t * pSrcA,
00072   q31_t * pSrcB,
00073   uint32_t blockSize,
00074   q63_t * result)
00075 {
00076   q63_t sum = 0;                                 /* Temporary result storage */
00077   uint32_t blkCnt;                               /* loop counter */
00078 
00079 
00080 #ifndef ARM_MATH_CM0_FAMILY
00081 
00082 /* Run the below code for Cortex-M4 and Cortex-M3 */
00083   q31_t inA1, inA2, inA3, inA4;
00084   q31_t inB1, inB2, inB3, inB4;
00085 
00086   /*loop Unrolling */
00087   blkCnt = blockSize >> 2u;
00088 
00089   /* First part of the processing with loop unrolling.  Compute 4 outputs at a time.    
00090    ** a second loop below computes the remaining 1 to 3 samples. */
00091   while(blkCnt > 0u)
00092   {
00093     /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */
00094     /* Calculate dot product and then store the result in a temporary buffer. */
00095     inA1 = *pSrcA++;
00096     inA2 = *pSrcA++;
00097     inA3 = *pSrcA++;
00098     inA4 = *pSrcA++;
00099     inB1 = *pSrcB++;
00100     inB2 = *pSrcB++;
00101     inB3 = *pSrcB++;
00102     inB4 = *pSrcB++;
00103 
00104     sum += ((q63_t) inA1 * inB1) >> 14u;
00105     sum += ((q63_t) inA2 * inB2) >> 14u;
00106     sum += ((q63_t) inA3 * inB3) >> 14u;
00107     sum += ((q63_t) inA4 * inB4) >> 14u;
00108 
00109     /* Decrement the loop counter */
00110     blkCnt--;
00111   }
00112 
00113   /* If the blockSize is not a multiple of 4, compute any remaining output samples here.    
00114    ** No loop unrolling is used. */
00115   blkCnt = blockSize % 0x4u;
00116 
00117 #else
00118 
00119   /* Run the below code for Cortex-M0 */
00120 
00121   /* Initialize blkCnt with number of samples */
00122   blkCnt = blockSize;
00123 
00124 #endif /* #ifndef ARM_MATH_CM0_FAMILY */
00125 
00126 
00127   while(blkCnt > 0u)
00128   {
00129     /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */
00130     /* Calculate dot product and then store the result in a temporary buffer. */
00131     sum += ((q63_t) * pSrcA++ * *pSrcB++) >> 14u;
00132 
00133     /* Decrement the loop counter */
00134     blkCnt--;
00135   }
00136 
00137   /* Store the result in the destination buffer in 16.48 format */
00138   *result = sum;
00139 }
00140 
00141 /**    
00142  * @} end of dot_prod group    
00143  */