Joep D / CMSIS_DSP_5

The CMSIS DSP 5 library

Dependents:   Nucleo-Heart-Rate ejercicioVrms2 PROYECTOFINAL ejercicioVrms ... more

functions/BasicMathFunctions/arm_dot_prod_q7.c@3:4098b9d3d571, 2018-06-21 (annotated)

Committer:
xorjoep
Date:
Thu Jun 21 11:56:27 2018 +0000
Revision:
3:4098b9d3d571
Parent:
1:24714b45cd1b 
headers is a folder not a library

Who changed what in which revision?

UserRevisionLine numberNew contents of line
xorjoep 1:24714b45cd1b 1 /* ----------------------------------------------------------------------
xorjoep 1:24714b45cd1b 2 * Project: CMSIS DSP Library
xorjoep 1:24714b45cd1b 3 * Title: arm_dot_prod_q7.c
xorjoep 1:24714b45cd1b 4 * Description: Q7 dot product
xorjoep 1:24714b45cd1b 5 *
xorjoep 1:24714b45cd1b 6 * $Date: 27. January 2017
xorjoep 1:24714b45cd1b 7 * $Revision: V.1.5.1
xorjoep 1:24714b45cd1b 8 *
xorjoep 1:24714b45cd1b 9 * Target Processor: Cortex-M cores
xorjoep 1:24714b45cd1b 10 * -------------------------------------------------------------------- */
xorjoep 1:24714b45cd1b 11 /*
xorjoep 1:24714b45cd1b 12 * Copyright (C) 2010-2017 ARM Limited or its affiliates. All rights reserved.
xorjoep 1:24714b45cd1b 13 *
xorjoep 1:24714b45cd1b 14 * SPDX-License-Identifier: Apache-2.0
xorjoep 1:24714b45cd1b 15 *
xorjoep 1:24714b45cd1b 16 * Licensed under the Apache License, Version 2.0 (the License); you may
xorjoep 1:24714b45cd1b 17 * not use this file except in compliance with the License.
xorjoep 1:24714b45cd1b 18 * You may obtain a copy of the License at
xorjoep 1:24714b45cd1b 19 *
xorjoep 1:24714b45cd1b 20 * www.apache.org/licenses/LICENSE-2.0
xorjoep 1:24714b45cd1b 21 *
xorjoep 1:24714b45cd1b 22 * Unless required by applicable law or agreed to in writing, software
xorjoep 1:24714b45cd1b 23 * distributed under the License is distributed on an AS IS BASIS, WITHOUT
xorjoep 1:24714b45cd1b 24 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
xorjoep 1:24714b45cd1b 25 * See the License for the specific language governing permissions and
xorjoep 1:24714b45cd1b 26 * limitations under the License.
xorjoep 1:24714b45cd1b 27 */
xorjoep 1:24714b45cd1b 28
xorjoep 1:24714b45cd1b 29 #include "arm_math.h"
xorjoep 1:24714b45cd1b 30
xorjoep 1:24714b45cd1b 31 /**
xorjoep 1:24714b45cd1b 32 * @ingroup groupMath
xorjoep 1:24714b45cd1b 33 */
xorjoep 1:24714b45cd1b 34
xorjoep 1:24714b45cd1b 35 /**
xorjoep 1:24714b45cd1b 36 * @addtogroup dot_prod
xorjoep 1:24714b45cd1b 37 * @{
xorjoep 1:24714b45cd1b 38 */
xorjoep 1:24714b45cd1b 39
xorjoep 1:24714b45cd1b 40 /**
xorjoep 1:24714b45cd1b 41 * @brief Dot product of Q7 vectors.
xorjoep 1:24714b45cd1b 42 * @param[in] *pSrcA points to the first input vector
xorjoep 1:24714b45cd1b 43 * @param[in] *pSrcB points to the second input vector
xorjoep 1:24714b45cd1b 44 * @param[in] blockSize number of samples in each vector
xorjoep 1:24714b45cd1b 45 * @param[out] *result output result returned here
xorjoep 1:24714b45cd1b 46 * @return none.
xorjoep 1:24714b45cd1b 47 *
xorjoep 1:24714b45cd1b 48 * <b>Scaling and Overflow Behavior:</b>
xorjoep 1:24714b45cd1b 49 * \par
xorjoep 1:24714b45cd1b 50 * The intermediate multiplications are in 1.7 x 1.7 = 2.14 format and these
xorjoep 1:24714b45cd1b 51 * results are added to an accumulator in 18.14 format.
xorjoep 1:24714b45cd1b 52 * Nonsaturating additions are used and there is no danger of wrap around as long as
xorjoep 1:24714b45cd1b 53 * the vectors are less than 2^18 elements long.
xorjoep 1:24714b45cd1b 54 * The return result is in 18.14 format.
xorjoep 1:24714b45cd1b 55 */
xorjoep 1:24714b45cd1b 56
xorjoep 1:24714b45cd1b 57 void arm_dot_prod_q7(
xorjoep 1:24714b45cd1b 58 q7_t * pSrcA,
xorjoep 1:24714b45cd1b 59 q7_t * pSrcB,
xorjoep 1:24714b45cd1b 60 uint32_t blockSize,
xorjoep 1:24714b45cd1b 61 q31_t * result)
xorjoep 1:24714b45cd1b 62 {
xorjoep 1:24714b45cd1b 63 uint32_t blkCnt; /* loop counter */
xorjoep 1:24714b45cd1b 64
xorjoep 1:24714b45cd1b 65 q31_t sum = 0; /* Temporary variables to store output */
xorjoep 1:24714b45cd1b 66
xorjoep 1:24714b45cd1b 67 #if defined (ARM_MATH_DSP)
xorjoep 1:24714b45cd1b 68
xorjoep 1:24714b45cd1b 69 /* Run the below code for Cortex-M4 and Cortex-M3 */
xorjoep 1:24714b45cd1b 70
xorjoep 1:24714b45cd1b 71 q31_t input1, input2; /* Temporary variables to store input */
xorjoep 1:24714b45cd1b 72 q31_t inA1, inA2, inB1, inB2; /* Temporary variables to store input */
xorjoep 1:24714b45cd1b 73
xorjoep 1:24714b45cd1b 74
xorjoep 1:24714b45cd1b 75
xorjoep 1:24714b45cd1b 76 /*loop Unrolling */
xorjoep 1:24714b45cd1b 77 blkCnt = blockSize >> 2U;
xorjoep 1:24714b45cd1b 78
xorjoep 1:24714b45cd1b 79 /* First part of the processing with loop unrolling. Compute 4 outputs at a time.
xorjoep 1:24714b45cd1b 80 ** a second loop below computes the remaining 1 to 3 samples. */
xorjoep 1:24714b45cd1b 81 while (blkCnt > 0U)
xorjoep 1:24714b45cd1b 82 {
xorjoep 1:24714b45cd1b 83 /* read 4 samples at a time from sourceA */
xorjoep 1:24714b45cd1b 84 input1 = *__SIMD32(pSrcA)++;
xorjoep 1:24714b45cd1b 85 /* read 4 samples at a time from sourceB */
xorjoep 1:24714b45cd1b 86 input2 = *__SIMD32(pSrcB)++;
xorjoep 1:24714b45cd1b 87
xorjoep 1:24714b45cd1b 88 /* extract two q7_t samples to q15_t samples */
xorjoep 1:24714b45cd1b 89 inA1 = __SXTB16(__ROR(input1, 8));
xorjoep 1:24714b45cd1b 90 /* extract reminaing two samples */
xorjoep 1:24714b45cd1b 91 inA2 = __SXTB16(input1);
xorjoep 1:24714b45cd1b 92 /* extract two q7_t samples to q15_t samples */
xorjoep 1:24714b45cd1b 93 inB1 = __SXTB16(__ROR(input2, 8));
xorjoep 1:24714b45cd1b 94 /* extract reminaing two samples */
xorjoep 1:24714b45cd1b 95 inB2 = __SXTB16(input2);
xorjoep 1:24714b45cd1b 96
xorjoep 1:24714b45cd1b 97 /* multiply and accumulate two samples at a time */
xorjoep 1:24714b45cd1b 98 sum = __SMLAD(inA1, inB1, sum);
xorjoep 1:24714b45cd1b 99 sum = __SMLAD(inA2, inB2, sum);
xorjoep 1:24714b45cd1b 100
xorjoep 1:24714b45cd1b 101 /* Decrement the loop counter */
xorjoep 1:24714b45cd1b 102 blkCnt--;
xorjoep 1:24714b45cd1b 103 }
xorjoep 1:24714b45cd1b 104
xorjoep 1:24714b45cd1b 105 /* If the blockSize is not a multiple of 4, compute any remaining output samples here.
xorjoep 1:24714b45cd1b 106 ** No loop unrolling is used. */
xorjoep 1:24714b45cd1b 107 blkCnt = blockSize % 0x4U;
xorjoep 1:24714b45cd1b 108
xorjoep 1:24714b45cd1b 109 while (blkCnt > 0U)
xorjoep 1:24714b45cd1b 110 {
xorjoep 1:24714b45cd1b 111 /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */
xorjoep 1:24714b45cd1b 112 /* Dot product and then store the results in a temporary buffer. */
xorjoep 1:24714b45cd1b 113 sum = __SMLAD(*pSrcA++, *pSrcB++, sum);
xorjoep 1:24714b45cd1b 114
xorjoep 1:24714b45cd1b 115 /* Decrement the loop counter */
xorjoep 1:24714b45cd1b 116 blkCnt--;
xorjoep 1:24714b45cd1b 117 }
xorjoep 1:24714b45cd1b 118
xorjoep 1:24714b45cd1b 119 #else
xorjoep 1:24714b45cd1b 120
xorjoep 1:24714b45cd1b 121 /* Run the below code for Cortex-M0 */
xorjoep 1:24714b45cd1b 122
xorjoep 1:24714b45cd1b 123
xorjoep 1:24714b45cd1b 124
xorjoep 1:24714b45cd1b 125 /* Initialize blkCnt with number of samples */
xorjoep 1:24714b45cd1b 126 blkCnt = blockSize;
xorjoep 1:24714b45cd1b 127
xorjoep 1:24714b45cd1b 128 while (blkCnt > 0U)
xorjoep 1:24714b45cd1b 129 {
xorjoep 1:24714b45cd1b 130 /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */
xorjoep 1:24714b45cd1b 131 /* Dot product and then store the results in a temporary buffer. */
xorjoep 1:24714b45cd1b 132 sum += (q31_t) ((q15_t) * pSrcA++ * *pSrcB++);
xorjoep 1:24714b45cd1b 133
xorjoep 1:24714b45cd1b 134 /* Decrement the loop counter */
xorjoep 1:24714b45cd1b 135 blkCnt--;
xorjoep 1:24714b45cd1b 136 }
xorjoep 1:24714b45cd1b 137
xorjoep 1:24714b45cd1b 138 #endif /* #if defined (ARM_MATH_DSP) */
xorjoep 1:24714b45cd1b 139
xorjoep 1:24714b45cd1b 140
xorjoep 1:24714b45cd1b 141 /* Store the result in the destination buffer in 18.14 format */
xorjoep 1:24714b45cd1b 142 *result = sum;
xorjoep 1:24714b45cd1b 143 }
xorjoep 1:24714b45cd1b 144
xorjoep 1:24714b45cd1b 145 /**
xorjoep 1:24714b45cd1b 146 * @} end of dot_prod group
xorjoep 1:24714b45cd1b 147 */