1 /* ----------------------------------------------------------------------
2 * Project: CMSIS DSP Library
3 * Title: arm_cmplx_mult_real_q15.c
4 * Description: Q15 complex by real multiplication
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 CmplxByRealMult
37 @{
38 */
39
40 /**
41 @brief Q15 complex-by-real multiplication.
42 @param[in] pSrcCmplx points to complex input vector
43 @param[in] pSrcReal points to real input vector
44 @param[out] pCmplxDst points to complex output vector
45 @param[in] numSamples number of samples in each vector
46 @return none
47
48 @par Scaling and Overflow Behavior
49 The function uses saturating arithmetic.
50 Results outside of the allowable Q15 range [0x8000 0x7FFF] are saturated.
51 */
52 #if defined(ARM_MATH_MVEI) && !defined(ARM_MATH_AUTOVECTORIZE)
53
arm_cmplx_mult_real_q15(const q15_t * pSrcCmplx,const q15_t * pSrcReal,q15_t * pCmplxDst,uint32_t numSamples)54 void arm_cmplx_mult_real_q15(
55 const q15_t * pSrcCmplx,
56 const q15_t * pSrcReal,
57 q15_t * pCmplxDst,
58 uint32_t numSamples)
59 {
60 static const uint16_t stride_cmplx_x_real_16[8] = {
61 0, 0, 1, 1, 2, 2, 3, 3
62 };
63 q15x8_t rVec;
64 q15x8_t cmplxVec;
65 q15x8_t dstVec;
66 uint16x8_t strideVec;
67 uint32_t blockSizeC = numSamples * CMPLX_DIM; /* loop counters */
68 uint32_t blkCnt;
69 q15_t in;
70
71 /*
72 * stride vector for pairs of real generation
73 */
74 strideVec = vld1q(stride_cmplx_x_real_16);
75
76 blkCnt = blockSizeC >> 3;
77
78 while (blkCnt > 0U)
79 {
80 cmplxVec = vld1q(pSrcCmplx);
81 rVec = vldrhq_gather_shifted_offset_s16(pSrcReal, strideVec);
82 dstVec = vqdmulhq(cmplxVec, rVec);
83 vst1q(pCmplxDst, dstVec);
84
85 pSrcReal += 4;
86 pSrcCmplx += 8;
87 pCmplxDst += 8;
88 blkCnt --;
89 }
90
91 /* Tail */
92 blkCnt = (blockSizeC & 7) >> 1;
93 while (blkCnt > 0U)
94 {
95 /* C[2 * i ] = A[2 * i ] * B[i]. */
96 /* C[2 * i + 1] = A[2 * i + 1] * B[i]. */
97
98 in = *pSrcReal++;
99 /* store the result in the destination buffer. */
100 *pCmplxDst++ = (q15_t) __SSAT((((q31_t) *pSrcCmplx++ * in) >> 15), 16);
101 *pCmplxDst++ = (q15_t) __SSAT((((q31_t) *pSrcCmplx++ * in) >> 15), 16);
102
103 /* Decrement loop counter */
104 blkCnt--;
105 }
106 }
107 #else
arm_cmplx_mult_real_q15(const q15_t * pSrcCmplx,const q15_t * pSrcReal,q15_t * pCmplxDst,uint32_t numSamples)108 void arm_cmplx_mult_real_q15(
109 const q15_t * pSrcCmplx,
110 const q15_t * pSrcReal,
111 q15_t * pCmplxDst,
112 uint32_t numSamples)
113 {
114 uint32_t blkCnt; /* Loop counter */
115 q15_t in; /* Temporary variable */
116
117 #if defined (ARM_MATH_LOOPUNROLL)
118
119 #if defined (ARM_MATH_DSP)
120 q31_t inA1, inA2; /* Temporary variables to hold input data */
121 q31_t inB1; /* Temporary variables to hold input data */
122 q15_t out1, out2, out3, out4; /* Temporary variables to hold output data */
123 q31_t mul1, mul2, mul3, mul4; /* Temporary variables to hold intermediate data */
124 #endif
125
126 /* Loop unrolling: Compute 4 outputs at a time */
127 blkCnt = numSamples >> 2U;
128
129 while (blkCnt > 0U)
130 {
131 /* C[2 * i ] = A[2 * i ] * B[i]. */
132 /* C[2 * i + 1] = A[2 * i + 1] * B[i]. */
133
134 #if defined (ARM_MATH_DSP)
135 /* read 2 complex numbers both real and imaginary from complex input buffer */
136 inA1 = read_q15x2_ia ((q15_t **) &pSrcCmplx);
137 inA2 = read_q15x2_ia ((q15_t **) &pSrcCmplx);
138 /* read 2 real values at a time from real input buffer */
139 inB1 = read_q15x2_ia ((q15_t **) &pSrcReal);
140
141 /* multiply complex number with real numbers */
142 #ifndef ARM_MATH_BIG_ENDIAN
143 mul1 = (q31_t) ((q15_t) (inA1) * (q15_t) (inB1));
144 mul2 = (q31_t) ((q15_t) (inA1 >> 16) * (q15_t) (inB1));
145 mul3 = (q31_t) ((q15_t) (inA2) * (q15_t) (inB1 >> 16));
146 mul4 = (q31_t) ((q15_t) (inA2 >> 16) * (q15_t) (inB1 >> 16));
147 #else
148 mul2 = (q31_t) ((q15_t) (inA1 >> 16) * (q15_t) (inB1 >> 16));
149 mul1 = (q31_t) ((q15_t) inA1 * (q15_t) (inB1 >> 16));
150 mul4 = (q31_t) ((q15_t) (inA2 >> 16) * (q15_t) inB1);
151 mul3 = (q31_t) ((q15_t) inA2 * (q15_t) inB1);
152 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
153
154 /* saturate the result */
155 out1 = (q15_t) __SSAT(mul1 >> 15U, 16);
156 out2 = (q15_t) __SSAT(mul2 >> 15U, 16);
157 out3 = (q15_t) __SSAT(mul3 >> 15U, 16);
158 out4 = (q15_t) __SSAT(mul4 >> 15U, 16);
159
160 /* pack real and imaginary outputs and store them to destination */
161 write_q15x2_ia (&pCmplxDst, __PKHBT(out1, out2, 16));
162 write_q15x2_ia (&pCmplxDst, __PKHBT(out3, out4, 16));
163
164 inA1 = read_q15x2_ia ((q15_t **) &pSrcCmplx);
165 inA2 = read_q15x2_ia ((q15_t **) &pSrcCmplx);
166 inB1 = read_q15x2_ia ((q15_t **) &pSrcReal);
167
168 #ifndef ARM_MATH_BIG_ENDIAN
169 mul1 = (q31_t) ((q15_t) (inA1) * (q15_t) (inB1));
170 mul2 = (q31_t) ((q15_t) (inA1 >> 16) * (q15_t) (inB1));
171 mul3 = (q31_t) ((q15_t) (inA2) * (q15_t) (inB1 >> 16));
172 mul4 = (q31_t) ((q15_t) (inA2 >> 16) * (q15_t) (inB1 >> 16));
173 #else
174 mul2 = (q31_t) ((q15_t) (inA1 >> 16) * (q15_t) (inB1 >> 16));
175 mul1 = (q31_t) ((q15_t) inA1 * (q15_t) (inB1 >> 16));
176 mul4 = (q31_t) ((q15_t) (inA2 >> 16) * (q15_t) inB1);
177 mul3 = (q31_t) ((q15_t) inA2 * (q15_t) inB1);
178 #endif /* #ifndef ARM_MATH_BIG_ENDIAN */
179
180 out1 = (q15_t) __SSAT(mul1 >> 15U, 16);
181 out2 = (q15_t) __SSAT(mul2 >> 15U, 16);
182 out3 = (q15_t) __SSAT(mul3 >> 15U, 16);
183 out4 = (q15_t) __SSAT(mul4 >> 15U, 16);
184
185 write_q15x2_ia (&pCmplxDst, __PKHBT(out1, out2, 16));
186 write_q15x2_ia (&pCmplxDst, __PKHBT(out3, out4, 16));
187 #else
188 in = *pSrcReal++;
189 *pCmplxDst++ = (q15_t) __SSAT((((q31_t) *pSrcCmplx++ * in) >> 15), 16);
190 *pCmplxDst++ = (q15_t) __SSAT((((q31_t) *pSrcCmplx++ * in) >> 15), 16);
191
192 in = *pSrcReal++;
193 *pCmplxDst++ = (q15_t) __SSAT((((q31_t) *pSrcCmplx++ * in) >> 15), 16);
194 *pCmplxDst++ = (q15_t) __SSAT((((q31_t) *pSrcCmplx++ * in) >> 15), 16);
195
196 in = *pSrcReal++;
197 *pCmplxDst++ = (q15_t) __SSAT((((q31_t) *pSrcCmplx++ * in) >> 15), 16);
198 *pCmplxDst++ = (q15_t) __SSAT((((q31_t) *pSrcCmplx++ * in) >> 15), 16);
199
200 in = *pSrcReal++;
201 *pCmplxDst++ = (q15_t) __SSAT((((q31_t) *pSrcCmplx++ * in) >> 15), 16);
202 *pCmplxDst++ = (q15_t) __SSAT((((q31_t) *pSrcCmplx++ * in) >> 15), 16);
203 #endif
204
205 /* Decrement loop counter */
206 blkCnt--;
207 }
208
209 /* Loop unrolling: Compute remaining outputs */
210 blkCnt = numSamples % 0x4U;
211
212 #else
213
214 /* Initialize blkCnt with number of samples */
215 blkCnt = numSamples;
216
217 #endif /* #if defined (ARM_MATH_LOOPUNROLL) */
218
219 while (blkCnt > 0U)
220 {
221 /* C[2 * i ] = A[2 * i ] * B[i]. */
222 /* C[2 * i + 1] = A[2 * i + 1] * B[i]. */
223
224 in = *pSrcReal++;
225 /* store the result in the destination buffer. */
226 *pCmplxDst++ = (q15_t) __SSAT((((q31_t) *pSrcCmplx++ * in) >> 15), 16);
227 *pCmplxDst++ = (q15_t) __SSAT((((q31_t) *pSrcCmplx++ * in) >> 15), 16);
228
229 /* Decrement loop counter */
230 blkCnt--;
231 }
232
233 }
234 #endif /* defined(ARM_MATH_MVEI) */
235
236 /**
237 @} end of CmplxByRealMult group
238 */
239