1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_offset_q7.c
4 * Description: Q7 vector offset
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 BasicOffset
37 @{
38 */
39
40 /**
41 @brief Adds a constant offset to a Q7 vector.
42 @param[in] pSrc points to the input vector
43 @param[in] offset is the offset to be added
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 Q7 range [0x80 0x7F] are saturated.
50 */
51 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
52
53 #include "arm_helium_utils.h"
54
arm_offset_q7(const q7_t * pSrc,q7_t offset,q7_t * pDst,uint32_t blockSize)55 ARM_DSP_ATTRIBUTE void arm_offset_q7(
56 const q7_t * pSrc,
57 q7_t offset,
58 q7_t * pDst,
59 uint32_t blockSize)
60 {
61 uint32_t blkCnt; /* loop counters */
62 q7x16_t vecSrc;
63
64 /* Compute 16 outputs at a time */
65 blkCnt = blockSize >> 4;
66 while (blkCnt > 0U)
67 {
68 /*
69 * C = A + offset
70 * Add offset and then store the result in the destination buffer.
71 */
72 vecSrc = vld1q(pSrc);
73 vst1q(pDst, vqaddq(vecSrc, offset));
74 /*
75 * Decrement the blockSize loop counter
76 */
77 blkCnt--;
78 /*
79 * advance vector source and destination pointers
80 */
81 pSrc += 16;
82 pDst += 16;
83 }
84 /*
85 * tail
86 */
87 blkCnt = blockSize & 0xF;
88 if (blkCnt > 0U)
89 {
90 mve_pred16_t p0 = vctp8q(blkCnt);
91 vecSrc = vld1q(pSrc);
92 vstrbq_p(pDst, vqaddq(vecSrc, offset), p0);
93 }
94 }
95
96 #else
arm_offset_q7(const q7_t * pSrc,q7_t offset,q7_t * pDst,uint32_t blockSize)97 ARM_DSP_ATTRIBUTE void arm_offset_q7(
98 const q7_t * pSrc,
99 q7_t offset,
100 q7_t * pDst,
101 uint32_t blockSize)
102 {
103 uint32_t blkCnt; /* Loop counter */
104
105 #if defined (ARM_MATH_LOOPUNROLL)
106
107 #if defined (ARM_MATH_DSP)
108 q31_t offset_packed; /* Offset packed to 32 bit */
109
110 /* Offset is packed to 32 bit in order to use SIMD32 for addition */
111 offset_packed = __PACKq7(offset, offset, offset, offset);
112 #endif
113
114 /* Loop unrolling: Compute 4 outputs at a time */
115 blkCnt = blockSize >> 2U;
116
117 while (blkCnt > 0U)
118 {
119 /* C = A + offset */
120
121 #if defined (ARM_MATH_DSP)
122 /* Add offset and store result in destination buffer (4 samples at a time). */
123 write_q7x4_ia (&pDst, __QADD8(read_q7x4_ia (&pSrc), offset_packed));
124 #else
125 *pDst++ = (q7_t) __SSAT((q15_t) *pSrc++ + offset, 8);
126 *pDst++ = (q7_t) __SSAT((q15_t) *pSrc++ + offset, 8);
127 *pDst++ = (q7_t) __SSAT((q15_t) *pSrc++ + offset, 8);
128 *pDst++ = (q7_t) __SSAT((q15_t) *pSrc++ + offset, 8);
129 #endif
130
131 /* Decrement loop counter */
132 blkCnt--;
133 }
134
135 /* Loop unrolling: Compute remaining outputs */
136 blkCnt = blockSize % 0x4U;
137
138 #else
139
140 /* Initialize blkCnt with number of samples */
141 blkCnt = blockSize;
142
143 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
144
145 while (blkCnt > 0U)
146 {
147 /* C = A + offset */
148
149 /* Add offset and store result in destination buffer. */
150 *pDst++ = (q7_t) __SSAT((q15_t) *pSrc++ + offset, 8);
151
152 /* Decrement loop counter */
153 blkCnt--;
154 }
155
156 }
157 #endif /* defined(ARM_MATH_MVEI) */
158
159 /**
160 @} end of BasicOffset group
161 */
162
