arm_dot_prod_q31.c
Go to the documentation of this file.
00001 /* ---------------------------------------------------------------------- 00002 * Copyright (C) 2011 ARM Limited. All rights reserved. 00003 * 00004 * $Date: 15. December 2011 00005 * $Revision: V2.0.0 00006 * 00007 * Project: Cortex-R DSP Library 00008 * Title: arm_dot_prod_q31.c 00009 * 00010 * Description: Q31 dot product. 00011 * 00012 * Target Processor: Cortex-R4/R5 00013 * 00014 * Version 1.0.0 2011/03/08 00015 * Alpha release. 00016 * 00017 * Version 1.0.1 2011/09/30 00018 * Beta release. 00019 * 00020 * Version 2.0.0 2011/12/15 00021 * Final release. 00022 * 00023 * -------------------------------------------------------------------- */ 00024 #include "arm_math.h" 00025 00053 void arm_dot_prod_q31( 00054 q31_t * pSrcA, 00055 q31_t * pSrcB, 00056 uint32_t blockSize, 00057 q63_t * result) 00058 { 00059 q63_t sum = 0, sum1 = 0; /* Temporary result storage */ 00060 uint32_t blkCnt; /* loop counter */ 00061 q31_t inA1, inA2, inB1, inB2; /* Temporary variables to store input data */ 00062 00063 /*loop Unrolling */ 00064 blkCnt = blockSize >> 3u; 00065 00066 /* First part of the processing with loop unrolling. Compute 8 outputs at a time. 00067 ** a second loop below computes the remaining 1 to 7 samples. */ 00068 while(blkCnt > 0u) 00069 { 00070 /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */ 00071 /* Calculate dot product and then store the result in a temporary buffer. */ 00072 00073 /* read input from sourceA buffer */ 00074 inA1 = *pSrcA; 00075 /* read input from sourceB buffer */ 00076 inB1 = *pSrcB; 00077 inA2 = *(pSrcA + 1); 00078 inB2 = *(pSrcB + 1); 00079 00080 /* multiply and accumulate in 16.48 format */ 00081 sum += ((q63_t)inA1 * inB1) >> 14u; 00082 sum1 += ((q63_t)inA2 * inB2) >> 14u; 00083 00084 inA1 = *(pSrcA + 2); 00085 inB1 = *(pSrcB + 2); 00086 inA2 = *(pSrcA + 3); 00087 inB2 = *(pSrcB + 3); 00088 /* multiply and accumulate in 16.48 format */ 00089 sum += ((q63_t)inA1 * inB1) >> 14u; 00090 sum1 += ((q63_t)inA2 * inB2) >> 14u; 00091 00092 inA1 = *(pSrcA + 4); 00093 inB1 = *(pSrcB + 4); 00094 inA2 = *(pSrcA + 5); 00095 inB2 = *(pSrcB + 5); 00096 /* multiply and accumulate in 16.48 format */ 00097 sum += ((q63_t)inA1 * inB1) >> 14u; 00098 sum1 += ((q63_t)inA2 * inB2) >> 14u; 00099 00100 inA1 = *(pSrcA + 6); 00101 inB1 = *(pSrcB + 6); 00102 inA2 = *(pSrcA + 7); 00103 inB2 = *(pSrcB + 7); 00104 /* multiply and accumulate in 16.48 format */ 00105 sum += ((q63_t)inA1 * inB1) >> 14u; 00106 sum1 += ((q63_t)inA2 * inB2) >> 14u; 00107 00108 /* increment sourceA pointer by 8 */ 00109 pSrcA += 8u; 00110 /* increment sourceB pointer bhy 8 */ 00111 pSrcB += 8u; 00112 00113 /* Decrement the loop counter */ 00114 blkCnt--; 00115 } 00116 sum = sum + sum1; 00117 /* If the blockSize is not a multiple of 8, compute any remaining output samples here. 00118 ** No loop unrolling is used. */ 00119 blkCnt = blockSize % 0x8u; 00120 00121 while(blkCnt > 0u) 00122 { 00123 /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */ 00124 /* Calculate dot product and then store the result in a temporary buffer. */ 00125 sum += ((q63_t) * pSrcA++ * *pSrcB++) >> 14u; 00126 00127 /* Decrement the loop counter */ 00128 blkCnt--; 00129 } 00130 00131 /* Store the result in the destination buffer in 16.48 format */ 00132 *result = sum; 00133 } 00134
