1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_q31_to_q7.c
4 * Description: Converts the elements of the Q31 vector to Q7 vector
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/support_functions.h"
30
31 /**
32 @ingroup groupSupport
33 */
34
35 /**
36 @addtogroup q31_to_x
37 @{
38 */
39
40 /**
41 @brief Converts the elements of the Q31 vector to Q7 vector.
42 @param[in] pSrc points to the Q31 input vector
43 @param[out] pDst points to the Q7 output vector
44 @param[in] blockSize number of samples in each vector
45
46 @par Details
47 The equation used for the conversion process is:
48 <pre>
49 pDst[n] = (q7_t) pSrc[n] >> 24; 0 <= n < blockSize.
50 </pre>
51 */
52 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
arm_q31_to_q7(const q31_t * pSrc,q7_t * pDst,uint32_t blockSize)53 ARM_DSP_ATTRIBUTE void arm_q31_to_q7(
54 const q31_t * pSrc,
55 q7_t * pDst,
56 uint32_t blockSize)
57 {
58 uint32_t blkCnt; /* loop counters */
59 q31x4x4_t tmp;
60 q15x8_t evVec = { 0 }, oddVec = { 0 };
61 q7x16_t vecDst = { 0 };
62 q31_t const *pSrcVec;
63
64 pSrcVec = (q31_t const *) pSrc;
65 blkCnt = blockSize >> 4;
66 while (blkCnt > 0U)
67 {
68 tmp = vld4q(pSrcVec);
69 pSrcVec += 16;
70 /* C = (q7_t) A >> 24 */
71 /* convert from q31 to q7 and then store the results in the destination buffer */
72 /*
73 * narrow and pack evens
74 */
75 evVec = vshrnbq_n_s32(evVec, tmp.val[0], 16);
76 evVec = vshrntq_n_s32(evVec, tmp.val[2], 16);
77 /*
78 * narrow and pack odds
79 */
80 oddVec = vshrnbq_n_s32(oddVec, tmp.val[1], 16);
81 oddVec = vshrntq_n_s32(oddVec, tmp.val[3], 16);
82 /*
83 * narrow & merge
84 */
85 vecDst = vshrnbq_n_s16(vecDst, evVec, 8);
86 vecDst = vshrntq_n_s16(vecDst, oddVec, 8);
87
88 vst1q(pDst, vecDst);
89 pDst += 16;
90 /*
91 * Decrement the blockSize loop counter
92 */
93 blkCnt--;
94 }
95 /*
96 * tail
97 */
98 blkCnt = blockSize & 0xF;
99 while (blkCnt > 0U)
100 {
101 /* C = (q7_t) (A >> 24) */
102
103 /* Convert from q31 to q7 and store result in destination buffer */
104 *pDst++ = (q7_t) (*pSrcVec++ >> 24);
105
106 /* Decrement loop counter */
107 blkCnt--;
108 }
109 }
110 #else
arm_q31_to_q7(const q31_t * pSrc,q7_t * pDst,uint32_t blockSize)111 ARM_DSP_ATTRIBUTE void arm_q31_to_q7(
112 const q31_t * pSrc,
113 q7_t * pDst,
114 uint32_t blockSize)
115 {
116 uint32_t blkCnt; /* Loop counter */
117 const q31_t *pIn = pSrc; /* Source pointer */
118
119 #if defined (ARM_MATH_LOOPUNROLL)
120
121 q7_t out1, out2, out3, out4;
122
123 /* Loop unrolling: Compute 4 outputs at a time */
124 blkCnt = blockSize >> 2U;
125
126 while (blkCnt > 0U)
127 {
128 /* C = (q7_t) (A >> 24) */
129
130 /* Convert from q31 to q7 and store result in destination buffer */
131
132 out1 = (q7_t) (*pIn++ >> 24);
133 out2 = (q7_t) (*pIn++ >> 24);
134 out3 = (q7_t) (*pIn++ >> 24);
135 out4 = (q7_t) (*pIn++ >> 24);
136 write_q7x4_ia (&pDst, __PACKq7(out1, out2, out3, out4));
137
138 /* Decrement loop counter */
139 blkCnt--;
140 }
141
142 /* Loop unrolling: Compute remaining outputs */
143 blkCnt = blockSize % 0x4U;
144
145 #else
146
147 /* Initialize blkCnt with number of samples */
148 blkCnt = blockSize;
149
150 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
151
152 while (blkCnt > 0U)
153 {
154 /* C = (q7_t) (A >> 24) */
155
156 /* Convert from q31 to q7 and store result in destination buffer */
157 *pDst++ = (q7_t) (*pIn++ >> 24);
158
159 /* Decrement loop counter */
160 blkCnt--;
161 }
162
163 }
164 #endif /* defined(ARM_MATH_MVEI) */
165
166 /**
167 @} end of q31_to_x group
168 */
169
