1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_shift_q31.c
4 * Description: Shifts the elements of a Q31 vector by a specified number of bits
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 @defgroup BasicShift Vector Shift
36
37 Shifts the elements of a fixed-point vector by a specified number of bits.
38 There are separate functions for Q7, Q15, and Q31 data types.
39 The underlying algorithm used is:
40
41 <pre>
42 pDst[n] = pSrc[n] << shift, 0 <= n < blockSize.
43 </pre>
44
45 If <code>shift</code> is positive then the elements of the vector are shifted to the left.
46 If <code>shift</code> is negative then the elements of the vector are shifted to the right.
47
48 The functions support in-place computation allowing the source and destination
49 pointers to reference the same memory buffer.
50 */
51
52 /**
53 @addtogroup BasicShift
54 @{
55 */
56
57 /**
58 @brief Shifts the elements of a Q31 vector a specified number of bits.
59 @param[in] pSrc points to the input vector
60 @param[in] shiftBits number of bits to shift. A positive value shifts left; a negative value shifts right.
61 @param[out] pDst points to the output vector
62 @param[in] blockSize number of samples in the vector
63 @return none
64
65 @par Scaling and Overflow Behavior
66 The function uses saturating arithmetic.
67 Results outside of the allowable Q31 range [0x80000000 0x7FFFFFFF] are saturated.
68 */
69
70 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
71
72 #include "arm_helium_utils.h"
73
arm_shift_q31(const q31_t * pSrc,int8_t shiftBits,q31_t * pDst,uint32_t blockSize)74 void arm_shift_q31(
75 const q31_t * pSrc,
76 int8_t shiftBits,
77 q31_t * pDst,
78 uint32_t blockSize)
79 {
80 uint32_t blkCnt; /* loop counters */
81 q31x4_t vecSrc;
82 q31x4_t vecDst;
83
84 /* Compute 4 outputs at a time */
85 blkCnt = blockSize >> 2;
86 while (blkCnt > 0U)
87 {
88 /*
89 * C = A (>> or <<) shiftBits
90 * Shift the input and then store the result in the destination buffer.
91 */
92 vecSrc = vld1q((q31_t const *) pSrc);
93 vecDst = vqshlq_r(vecSrc, shiftBits);
94 vst1q(pDst, vecDst);
95 /*
96 * Decrement the blockSize loop counter
97 */
98 blkCnt--;
99 /*
100 * advance vector source and destination pointers
101 */
102 pSrc += 4;
103 pDst += 4;
104 }
105 /*
106 * tail
107 */
108 blkCnt = blockSize & 3;
109 if (blkCnt > 0U)
110 {
111 mve_pred16_t p0 = vctp32q(blkCnt);
112 vecSrc = vld1q((q31_t const *) pSrc);
113 vecDst = vqshlq_r(vecSrc, shiftBits);
114 vstrwq_p(pDst, vecDst, p0);
115 }
116 }
117
118
119 #else
arm_shift_q31(const q31_t * pSrc,int8_t shiftBits,q31_t * pDst,uint32_t blockSize)120 void arm_shift_q31(
121 const q31_t * pSrc,
122 int8_t shiftBits,
123 q31_t * pDst,
124 uint32_t blockSize)
125 {
126 uint32_t blkCnt; /* Loop counter */
127 uint8_t sign = (shiftBits & 0x80); /* Sign of shiftBits */
128
129 #if defined (ARM_MATH_LOOPUNROLL)
130
131 q31_t in, out; /* Temporary variables */
132
133 /* Loop unrolling: Compute 4 outputs at a time */
134 blkCnt = blockSize >> 2U;
135
136 /* If the shift value is positive then do right shift else left shift */
137 if (sign == 0U)
138 {
139 while (blkCnt > 0U)
140 {
141 /* C = A << shiftBits */
142
143 /* Shift input and store result in destination buffer. */
144 in = *pSrc++;
145 out = in << shiftBits;
146 if (in != (out >> shiftBits))
147 out = 0x7FFFFFFF ^ (in >> 31);
148 *pDst++ = out;
149
150 in = *pSrc++;
151 out = in << shiftBits;
152 if (in != (out >> shiftBits))
153 out = 0x7FFFFFFF ^ (in >> 31);
154 *pDst++ = out;
155
156 in = *pSrc++;
157 out = in << shiftBits;
158 if (in != (out >> shiftBits))
159 out = 0x7FFFFFFF ^ (in >> 31);
160 *pDst++ = out;
161
162 in = *pSrc++;
163 out = in << shiftBits;
164 if (in != (out >> shiftBits))
165 out = 0x7FFFFFFF ^ (in >> 31);
166 *pDst++ = out;
167
168 /* Decrement loop counter */
169 blkCnt--;
170 }
171 }
172 else
173 {
174 while (blkCnt > 0U)
175 {
176 /* C = A >> shiftBits */
177
178 /* Shift input and store results in destination buffer. */
179 *pDst++ = (*pSrc++ >> -shiftBits);
180 *pDst++ = (*pSrc++ >> -shiftBits);
181 *pDst++ = (*pSrc++ >> -shiftBits);
182 *pDst++ = (*pSrc++ >> -shiftBits);
183
184 /* Decrement loop counter */
185 blkCnt--;
186 }
187 }
188
189 /* Loop unrolling: Compute remaining outputs */
190 blkCnt = blockSize % 0x4U;
191
192 #else
193
194 /* Initialize blkCnt with number of samples */
195 blkCnt = blockSize;
196
197 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
198
199 /* If the shift value is positive then do right shift else left shift */
200 if (sign == 0U)
201 {
202 while (blkCnt > 0U)
203 {
204 /* C = A << shiftBits */
205
206 /* Shift input and store result in destination buffer. */
207 *pDst++ = clip_q63_to_q31((q63_t) *pSrc++ << shiftBits);
208
209 /* Decrement loop counter */
210 blkCnt--;
211 }
212 }
213 else
214 {
215 while (blkCnt > 0U)
216 {
217 /* C = A >> shiftBits */
218
219 /* Shift input and store result in destination buffer. */
220 *pDst++ = (*pSrc++ >> -shiftBits);
221
222 /* Decrement loop counter */
223 blkCnt--;
224 }
225 }
226
227 }
228 #endif /* defined(ARM_MATH_MVEI) */
229
230 /**
231 @} end of BasicShift group
232 */
233
