1 /* ----------------------------------------------------------------------
2  * Project:      CMSIS DSP Library
3  * Title:        arm_add_q15.c
4  * Description:  Q15 vector addition
5  *
6  * $Date:        23 April 2021
7  * $Revision:    V1.9.0
8  *
9  * Target Processor: Cortex-M and Cortex-A cores
10  * -------------------------------------------------------------------- */
11 /*
12  * Copyright (C) 2010-2021 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 "dsp/basic_math_functions.h"
30 
31 /**
32   @ingroup groupMath
33  */
34 
35 /**
36   @addtogroup BasicAdd
37   @{
38  */
39 
40 /**
41   @brief         Q15 vector addition.
42   @param[in]     pSrcA      points to the first input vector
43   @param[in]     pSrcB      points to the second input vector
44   @param[out]    pDst       points to the output vector
45   @param[in]     blockSize  number of samples in each vector
46 
47   @par           Scaling and Overflow Behavior
48                    The function uses saturating arithmetic.
49                    Results outside of the allowable Q15 range [0x8000 0x7FFF] are saturated.
50  */
51 
52 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
53 
54 #include "arm_helium_utils.h"
55 
arm_add_q15(const q15_t * pSrcA,const q15_t * pSrcB,q15_t * pDst,uint32_t blockSize)56 void arm_add_q15(
57     const q15_t * pSrcA,
58     const q15_t * pSrcB,
59     q15_t * pDst,
60     uint32_t blockSize)
61 {
62     uint32_t  blkCnt;           /* loop counters */
63     q15x8_t vecA;
64     q15x8_t vecB;
65 
66     /* Compute 8 outputs at a time */
67     blkCnt = blockSize >> 3;
68     while (blkCnt > 0U)
69     {
70         /*
71          * C = A + B
72          * Add and then store the results in the destination buffer.
73          */
74         vecA = vld1q(pSrcA);
75         vecB = vld1q(pSrcB);
76         vst1q(pDst, vqaddq(vecA, vecB));
77         /*
78          * Decrement the blockSize loop counter
79          */
80         blkCnt--;
81         /*
82          * advance vector source and destination pointers
83          */
84         pSrcA  += 8;
85         pSrcB  += 8;
86         pDst   += 8;
87     }
88     /*
89      * tail
90      */
91     blkCnt = blockSize & 7;
92     if (blkCnt > 0U)
93     {
94         mve_pred16_t p0 = vctp16q(blkCnt);
95         vecA = vld1q(pSrcA);
96         vecB = vld1q(pSrcB);
97         vstrhq_p(pDst, vqaddq(vecA, vecB), p0);
98     }
99 }
100 
101 #else
arm_add_q15(const q15_t * pSrcA,const q15_t * pSrcB,q15_t * pDst,uint32_t blockSize)102 void arm_add_q15(
103   const q15_t * pSrcA,
104   const q15_t * pSrcB,
105         q15_t * pDst,
106         uint32_t blockSize)
107 {
108         uint32_t blkCnt;                               /* Loop counter */
109 
110 #if defined (ARM_MATH_LOOPUNROLL)
111 
112 #if defined (ARM_MATH_DSP)
113   q31_t inA1, inA2;
114   q31_t inB1, inB2;
115 #endif
116 
117   /* Loop unrolling: Compute 4 outputs at a time */
118   blkCnt = blockSize >> 2U;
119 
120   while (blkCnt > 0U)
121   {
122     /* C = A + B */
123 
124 #if defined (ARM_MATH_DSP)
125     /* read 2 times 2 samples at a time from sourceA */
126     inA1 = read_q15x2_ia (&pSrcA);
127     inA2 = read_q15x2_ia (&pSrcA);
128     /* read 2 times 2 samples at a time from sourceB */
129     inB1 = read_q15x2_ia (&pSrcB);
130     inB2 = read_q15x2_ia (&pSrcB);
131 
132     /* Add and store 2 times 2 samples at a time */
133     write_q15x2_ia (&pDst, __QADD16(inA1, inB1));
134     write_q15x2_ia (&pDst, __QADD16(inA2, inB2));
135 #else
136     *pDst++ = (q15_t) __SSAT(((q31_t) *pSrcA++ + *pSrcB++), 16);
137     *pDst++ = (q15_t) __SSAT(((q31_t) *pSrcA++ + *pSrcB++), 16);
138     *pDst++ = (q15_t) __SSAT(((q31_t) *pSrcA++ + *pSrcB++), 16);
139     *pDst++ = (q15_t) __SSAT(((q31_t) *pSrcA++ + *pSrcB++), 16);
140 #endif
141 
142     /* Decrement loop counter */
143     blkCnt--;
144   }
145 
146   /* Loop unrolling: Compute remaining outputs */
147   blkCnt = blockSize % 0x4U;
148 
149 #else
150 
151   /* Initialize blkCnt with number of samples */
152   blkCnt = blockSize;
153 
154 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
155 
156   while (blkCnt > 0U)
157   {
158     /* C = A + B */
159 
160     /* Add and store result in destination buffer. */
161 #if defined (ARM_MATH_DSP)
162     *pDst++ = (q15_t) __QADD16(*pSrcA++, *pSrcB++);
163 #else
164     *pDst++ = (q15_t) __SSAT(((q31_t) *pSrcA++ + *pSrcB++), 16);
165 #endif
166 
167     /* Decrement loop counter */
168     blkCnt--;
169   }
170 
171 }
172 #endif /* defined(ARM_MATH_MVEI) */
173 /**
174   @} end of BasicAdd group
175  */
176