1 /*
2 * Copyright (C) 2010-2018 Arm Limited or its affiliates. All rights reserved.
3 *
4 * SPDX-License-Identifier: Apache-2.0
5 *
6 * Licensed under the Apache License, Version 2.0 (the License); you may
7 * not use this file except in compliance with the License.
8 * You may obtain a copy of the License at
9 *
10 * www.apache.org/licenses/LICENSE-2.0
11 *
12 * Unless required by applicable law or agreed to in writing, software
13 * distributed under the License is distributed on an AS IS BASIS, WITHOUT
14 * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
15 * See the License for the specific language governing permissions and
16 * limitations under the License.
17 */
18
19 /* ----------------------------------------------------------------------
20 * Project: CMSIS NN Library
21 * Title: arm_nn_mult_q7.c
22 * Description: Q7 vector multiplication with variable output shifts
23 *
24 * $Date: 09. October 2020
25 * $Revision: V.1.0.2
26 *
27 * Target Processor: Cortex-M cores
28 *
29 * -------------------------------------------------------------------- */
30
31 #include "arm_nnsupportfunctions.h"
32
33 /**
34 * @ingroup groupSupport
35 */
36
37 /**
38 * @addtogroup NNBasicMath
39 * @{
40 */
41
42 /**
43 * @brief Q7 vector multiplication with variable output shifts
44 * @param[in] *pSrcA pointer to the first input vector
45 * @param[in] *pSrcB pointer to the second input vector
46 * @param[out] *pDst pointer to the output vector
47 * @param[in] out_shift amount of right-shift for output
48 * @param[in] blockSize number of samples in each vector
49 *
50 * <b>Scaling and Overflow Behavior:</b>
51 * \par
52 * The function uses saturating arithmetic.
53 * Results outside of the allowable Q7 range [0x80 0x7F] will be saturated.
54 */
55
arm_nn_mult_q7(q7_t * pSrcA,q7_t * pSrcB,q7_t * pDst,const uint16_t out_shift,uint32_t blockSize)56 void arm_nn_mult_q7(q7_t *pSrcA, q7_t *pSrcB, q7_t *pDst, const uint16_t out_shift, uint32_t blockSize)
57 {
58 uint32_t blkCnt; /* loop counters */
59
60 #if defined(ARM_MATH_DSP)
61
62 /* Run the below code for Cortex-M4 and Cortex-M3 */
63 q7_t out1, out2, out3, out4; /* Temporary variables to store the product */
64
65 /* loop Unrolling */
66 blkCnt = blockSize >> 2U;
67
68 /* First part of the processing with loop unrolling. Compute 4 outputs at a time.
69 ** a second loop below computes the remaining 1 to 3 samples. */
70 while (blkCnt > 0U)
71 {
72 /* C = A * B */
73 /* Multiply the inputs and store the results in temporary variables */
74 out1 = (q7_t)__SSAT(((q15_t)((q15_t)(*pSrcA++) * (*pSrcB++) + NN_ROUND(out_shift)) >> out_shift), 8);
75 out2 = (q7_t)__SSAT(((q15_t)((q15_t)(*pSrcA++) * (*pSrcB++) + NN_ROUND(out_shift)) >> out_shift), 8);
76 out3 = (q7_t)__SSAT(((q15_t)((q15_t)(*pSrcA++) * (*pSrcB++) + NN_ROUND(out_shift)) >> out_shift), 8);
77 out4 = (q7_t)__SSAT(((q15_t)((q15_t)(*pSrcA++) * (*pSrcB++) + NN_ROUND(out_shift)) >> out_shift), 8);
78
79 /* Store the results of 4 inputs in the destination buffer in single cycle by packing */
80 *__SIMD32(pDst)++ = __PACKq7(out1, out2, out3, out4);
81
82 /* Decrement the blockSize loop counter */
83 blkCnt--;
84 }
85
86 /* If the blockSize is not a multiple of 4, compute any remaining output samples here.
87 ** No loop unrolling is used. */
88 blkCnt = blockSize % 0x4U;
89
90 #else
91
92 /* Run the below code for Cortex-M0 */
93
94 /* Initialize blkCnt with number of samples */
95 blkCnt = blockSize;
96
97 #endif /* #if defined (ARM_MATH_DSP) */
98
99 while (blkCnt > 0U)
100 {
101 /* C = A * B */
102 /* Multiply the inputs and store the result in the destination buffer */
103 *pDst++ = (q7_t)__SSAT(((q15_t)((q15_t)(*pSrcA++) * (*pSrcB++) + NN_ROUND(out_shift)) >> out_shift), 8);
104
105 /* Decrement the blockSize loop counter */
106 blkCnt--;
107 }
108 }
109
110 /**
111 * @} end of NNBasicMath group
112 */
113
