1 /* ----------------------------------------------------------------------
2  * Project:      CMSIS DSP Library
3  * Title:        arm_cmplx_mag_q15.c
4  * Description:  Q15 complex magnitude
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/complex_math_functions.h"
30 
31 /**
32   @ingroup groupCmplxMath
33  */
34 
35 /**
36   @addtogroup cmplx_mag
37   @{
38  */
39 
40 /**
41   @brief         Q15 complex magnitude.
42   @param[in]     pSrc        points to input vector
43   @param[out]    pDst        points to output vector
44   @param[in]     numSamples  number of samples in each vector
45   @return        none
46 
47   @par           Scaling and Overflow Behavior
48                    The function implements 1.15 by 1.15 multiplications and finally output is converted into 2.14 format.
49  */
50 
51 /* Sqrt q31 is used otherwise accuracy is not good enough
52            for small values and for some applications it is
53            an issue.
54         */
55 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
56 
57 #include "arm_helium_utils.h"
58 
arm_cmplx_mag_q15(const q15_t * pSrc,q15_t * pDst,uint32_t numSamples)59 void arm_cmplx_mag_q15(
60   const q15_t * pSrc,
61         q15_t * pDst,
62         uint32_t numSamples)
63 {
64 
65     int32_t blockSize = numSamples;  /* loop counters */
66     uint32_t  blkCnt;           /* loop counters */
67     q15x8x2_t vecSrc;
68     q31x4_t prod0;
69     q31x4_t prod1;
70 
71     q31_t in;
72     q31_t acc0;
73     q31x4_t acc0V;
74     q31x4_t acc1V;
75 
76     q31_t res;
77     q15x8_t resV;
78 
79     blkCnt = blockSize >> 3;
80     while (blkCnt > 0U)
81     {
82         vecSrc = vld2q(pSrc);
83         pSrc += 16;
84 
85         acc0V = vdupq_n_s32(0);
86         acc1V = vdupq_n_s32(0);
87 
88         prod0 = vmullbq_int_s16(vecSrc.val[0], vecSrc.val[0]);
89         acc0V = vqaddq_s32(acc0V,prod0);
90 
91         prod0 = vmullbq_int_s16(vecSrc.val[1], vecSrc.val[1]);
92         acc0V = vqaddq_s32(acc0V,prod0);
93 
94 
95         prod1 = vmulltq_int_s16(vecSrc.val[0], vecSrc.val[0]);
96         acc1V = vqaddq_s32(acc1V,prod1);
97 
98         prod1 = vmulltq_int_s16(vecSrc.val[1], vecSrc.val[1]);
99         acc1V = vqaddq_s32(acc1V,prod1);
100 
101 
102 
103         acc0V = vshrq(acc0V, 1);
104         acc1V = vshrq(acc1V, 1);
105 
106         acc0V = FAST_VSQRT_Q31(acc0V);
107         acc1V = FAST_VSQRT_Q31(acc1V);
108 
109         resV = vdupq_n_s16(0);
110         resV = vqshrnbq_n_s32(resV,acc0V,16);
111         resV = vqshrntq_n_s32(resV,acc1V,16);
112 
113         vst1q(pDst, resV);
114         pDst += 8;
115         /*
116          * Decrement the blockSize loop counter
117          */
118         blkCnt--;
119     }
120 
121     /*
122      * tail
123      */
124     blkCnt = blockSize & 7;
125 
126     while (blkCnt > 0U)
127     {
128       /* C[0] = sqrt(A[0] * A[0] + A[1] * A[1]) */
129 
130       in = read_q15x2_ia ((q15_t **) &pSrc);
131       acc0 = __SMUAD(in, in);
132 
133       /* store result in 2.14 format in destination buffer. */
134       arm_sqrt_q31((uint32_t)acc0  >> 1 , &res);
135       *pDst++ = res >> 16;
136 
137 
138       /* Decrement loop counter */
139       blkCnt--;
140     }
141 }
142 
143 #else
arm_cmplx_mag_q15(const q15_t * pSrc,q15_t * pDst,uint32_t numSamples)144 void arm_cmplx_mag_q15(
145   const q15_t * pSrc,
146         q15_t * pDst,
147         uint32_t numSamples)
148 {
149         q31_t res; /* temporary result */
150         uint32_t blkCnt;                               /* Loop counter */
151 
152 #if defined (ARM_MATH_DSP)
153         q31_t in;
154         q31_t acc0;                                    /* Accumulators */
155 #else
156        q15_t real, imag;                              /* Temporary input variables */
157        q31_t acc0, acc1;                              /* Accumulators */
158 #endif
159 
160 #if defined (ARM_MATH_LOOPUNROLL)
161 
162   /* Loop unrolling: Compute 4 outputs at a time */
163   blkCnt = numSamples >> 2U;
164 
165   while (blkCnt > 0U)
166   {
167     /* C[0] = sqrt(A[0] * A[0] + A[1] * A[1]) */
168 
169 #if defined (ARM_MATH_DSP)
170     in = read_q15x2_ia (&pSrc);
171     acc0 = __SMUAD(in, in);
172     /* store result in 2.14 format in destination buffer. */
173     arm_sqrt_q31((uint32_t)acc0  >> 1 , &res);
174     *pDst++ = res >> 16;
175 
176     in = read_q15x2_ia (&pSrc);
177     acc0 = __SMUAD(in, in);
178     arm_sqrt_q31((uint32_t)acc0  >> 1 , &res);
179     *pDst++ = res >> 16;
180 
181     in = read_q15x2_ia (&pSrc);
182     acc0 = __SMUAD(in, in);
183     arm_sqrt_q31((uint32_t)acc0  >> 1 , &res);
184     *pDst++ = res >> 16;
185 
186     in = read_q15x2_ia (&pSrc);
187     acc0 = __SMUAD(in, in);
188     arm_sqrt_q31((uint32_t)acc0  >> 1 , &res);
189     *pDst++ = res >> 16;
190 #else
191     real = *pSrc++;
192     imag = *pSrc++;
193     acc0 = ((q31_t) real * real);
194     acc1 = ((q31_t) imag * imag);
195 
196     /* store result in 2.14 format in destination buffer. */
197     arm_sqrt_q31(((uint32_t)acc0 + (uint32_t)acc1) >> 1 , &res);
198     *pDst++ = res >> 16;
199 
200     real = *pSrc++;
201     imag = *pSrc++;
202     acc0 = ((q31_t) real * real);
203     acc1 = ((q31_t) imag * imag);
204     arm_sqrt_q31(((uint32_t)acc0 + (uint32_t)acc1) >> 1 , &res);
205     *pDst++ = res >> 16;
206 
207     real = *pSrc++;
208     imag = *pSrc++;
209     acc0 = ((q31_t) real * real);
210     acc1 = ((q31_t) imag * imag);
211     arm_sqrt_q31(((uint32_t)acc0 + (uint32_t)acc1) >> 1 , &res);
212     *pDst++ = res >> 16;
213 
214     real = *pSrc++;
215     imag = *pSrc++;
216     acc0 = ((q31_t) real * real);
217     acc1 = ((q31_t) imag * imag);
218     arm_sqrt_q31(((uint32_t)acc0 + (uint32_t)acc1) >> 1 , &res);
219     *pDst++ = res >> 16;
220 #endif /* #if defined (ARM_MATH_DSP) */
221 
222     /* Decrement loop counter */
223     blkCnt--;
224   }
225 
226   /* Loop unrolling: Compute remaining outputs */
227   blkCnt = numSamples % 0x4U;
228 
229 #else
230 
231   /* Initialize blkCnt with number of samples */
232   blkCnt = numSamples;
233 
234 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
235 
236   while (blkCnt > 0U)
237   {
238     /* C[0] = sqrt(A[0] * A[0] + A[1] * A[1]) */
239 
240 #if defined (ARM_MATH_DSP)
241     in = read_q15x2_ia (&pSrc);
242     acc0 = __SMUAD(in, in);
243     /* store result in 2.14 format in destination buffer. */
244     arm_sqrt_q31((uint32_t)acc0  >> 1 , &res);
245     *pDst++ = res >> 16;
246 #else
247     real = *pSrc++;
248     imag = *pSrc++;
249     acc0 = ((q31_t) real * real);
250     acc1 = ((q31_t) imag * imag);
251 
252     /* store result in 2.14 format in destination buffer. */
253     arm_sqrt_q31(((uint32_t)acc0 + (uint32_t)acc1) >> 1 , &res);
254     *pDst++ = res >> 16;
255 
256 #endif
257 
258     /* Decrement loop counter */
259     blkCnt--;
260   }
261 
262 }
263 #endif /* defined(ARM_MATH_MVEI) */
264 
265 /**
266   @} end of cmplx_mag group
267  */
268