Details | Last modification | View Log | RSS feed
| Rev | Author | Line No. | Line |
|---|---|---|---|
| 2 | mjames | 1 | /* ---------------------------------------------------------------------- |
| 2 | * Project: CMSIS DSP Library |
||
| 3 | * Title: arm_dot_prod_q31.c |
||
| 4 | * Description: Q31 dot product |
||
| 5 | * |
||
| 6 | * $Date: 27. January 2017 |
||
| 7 | * $Revision: V.1.5.1 |
||
| 8 | * |
||
| 9 | * Target Processor: Cortex-M cores |
||
| 10 | * -------------------------------------------------------------------- */ |
||
| 11 | /* |
||
| 12 | * Copyright (C) 2010-2017 ARM Limited or its affiliates. All rights reserved. |
||
| 13 | * |
||
| 14 | * SPDX-License-Identifier: Apache-2.0 |
||
| 15 | * |
||
| 16 | * Licensed under the Apache License, Version 2.0 (the License); you may |
||
| 17 | * not use this file except in compliance with the License. |
||
| 18 | * You may obtain a copy of the License at |
||
| 19 | * |
||
| 20 | * www.apache.org/licenses/LICENSE-2.0 |
||
| 21 | * |
||
| 22 | * Unless required by applicable law or agreed to in writing, software |
||
| 23 | * distributed under the License is distributed on an AS IS BASIS, WITHOUT |
||
| 24 | * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
||
| 25 | * See the License for the specific language governing permissions and |
||
| 26 | * limitations under the License. |
||
| 27 | */ |
||
| 28 | |||
| 29 | #include "arm_math.h" |
||
| 30 | |||
| 31 | /** |
||
| 32 | * @ingroup groupMath |
||
| 33 | */ |
||
| 34 | |||
| 35 | /** |
||
| 36 | * @addtogroup dot_prod |
||
| 37 | * @{ |
||
| 38 | */ |
||
| 39 | |||
| 40 | /** |
||
| 41 | * @brief Dot product of Q31 vectors. |
||
| 42 | * @param[in] *pSrcA points to the first input vector |
||
| 43 | * @param[in] *pSrcB points to the second input vector |
||
| 44 | * @param[in] blockSize number of samples in each vector |
||
| 45 | * @param[out] *result output result returned here |
||
| 46 | * @return none. |
||
| 47 | * |
||
| 48 | * <b>Scaling and Overflow Behavior:</b> |
||
| 49 | * \par |
||
| 50 | * The intermediate multiplications are in 1.31 x 1.31 = 2.62 format and these |
||
| 51 | * are truncated to 2.48 format by discarding the lower 14 bits. |
||
| 52 | * The 2.48 result is then added without saturation to a 64-bit accumulator in 16.48 format. |
||
| 53 | * There are 15 guard bits in the accumulator and there is no risk of overflow as long as |
||
| 54 | * the length of the vectors is less than 2^16 elements. |
||
| 55 | * The return result is in 16.48 format. |
||
| 56 | */ |
||
| 57 | |||
| 58 | void arm_dot_prod_q31( |
||
| 59 | q31_t * pSrcA, |
||
| 60 | q31_t * pSrcB, |
||
| 61 | uint32_t blockSize, |
||
| 62 | q63_t * result) |
||
| 63 | { |
||
| 64 | q63_t sum = 0; /* Temporary result storage */ |
||
| 65 | uint32_t blkCnt; /* loop counter */ |
||
| 66 | |||
| 67 | |||
| 68 | #if defined (ARM_MATH_DSP) |
||
| 69 | |||
| 70 | /* Run the below code for Cortex-M4 and Cortex-M3 */ |
||
| 71 | q31_t inA1, inA2, inA3, inA4; |
||
| 72 | q31_t inB1, inB2, inB3, inB4; |
||
| 73 | |||
| 74 | /*loop Unrolling */ |
||
| 75 | blkCnt = blockSize >> 2U; |
||
| 76 | |||
| 77 | /* First part of the processing with loop unrolling. Compute 4 outputs at a time. |
||
| 78 | ** a second loop below computes the remaining 1 to 3 samples. */ |
||
| 79 | while (blkCnt > 0U) |
||
| 80 | { |
||
| 81 | /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */ |
||
| 82 | /* Calculate dot product and then store the result in a temporary buffer. */ |
||
| 83 | inA1 = *pSrcA++; |
||
| 84 | inA2 = *pSrcA++; |
||
| 85 | inA3 = *pSrcA++; |
||
| 86 | inA4 = *pSrcA++; |
||
| 87 | inB1 = *pSrcB++; |
||
| 88 | inB2 = *pSrcB++; |
||
| 89 | inB3 = *pSrcB++; |
||
| 90 | inB4 = *pSrcB++; |
||
| 91 | |||
| 92 | sum += ((q63_t) inA1 * inB1) >> 14U; |
||
| 93 | sum += ((q63_t) inA2 * inB2) >> 14U; |
||
| 94 | sum += ((q63_t) inA3 * inB3) >> 14U; |
||
| 95 | sum += ((q63_t) inA4 * inB4) >> 14U; |
||
| 96 | |||
| 97 | /* Decrement the loop counter */ |
||
| 98 | blkCnt--; |
||
| 99 | } |
||
| 100 | |||
| 101 | /* If the blockSize is not a multiple of 4, compute any remaining output samples here. |
||
| 102 | ** No loop unrolling is used. */ |
||
| 103 | blkCnt = blockSize % 0x4U; |
||
| 104 | |||
| 105 | #else |
||
| 106 | |||
| 107 | /* Run the below code for Cortex-M0 */ |
||
| 108 | |||
| 109 | /* Initialize blkCnt with number of samples */ |
||
| 110 | blkCnt = blockSize; |
||
| 111 | |||
| 112 | #endif /* #if defined (ARM_MATH_DSP) */ |
||
| 113 | |||
| 114 | |||
| 115 | while (blkCnt > 0U) |
||
| 116 | { |
||
| 117 | /* C = A[0]* B[0] + A[1]* B[1] + A[2]* B[2] + .....+ A[blockSize-1]* B[blockSize-1] */ |
||
| 118 | /* Calculate dot product and then store the result in a temporary buffer. */ |
||
| 119 | sum += ((q63_t) * pSrcA++ * *pSrcB++) >> 14U; |
||
| 120 | |||
| 121 | /* Decrement the loop counter */ |
||
| 122 | blkCnt--; |
||
| 123 | } |
||
| 124 | |||
| 125 | /* Store the result in the destination buffer in 16.48 format */ |
||
| 126 | *result = sum; |
||
| 127 | } |
||
| 128 | |||
| 129 | /** |
||
| 130 | * @} end of dot_prod group |
||
| 131 | */ |