Line | Branch | Exec | Source |
---|---|---|---|
1 | /* -*- c++ -*- */ | ||
2 | /* | ||
3 | * Copyright 2012, 2014 Free Software Foundation, Inc. | ||
4 | * | ||
5 | * This file is part of VOLK | ||
6 | * | ||
7 | * SPDX-License-Identifier: LGPL-3.0-or-later | ||
8 | */ | ||
9 | |||
10 | /*! | ||
11 | * \page volk_32fc_deinterleave_real_32f | ||
12 | * | ||
13 | * \b Overview | ||
14 | * | ||
15 | * Deinterleaves the complex floating point vector and return the real | ||
16 | * part (inphase) of the samples. | ||
17 | * | ||
18 | * <b>Dispatcher Prototype</b> | ||
19 | * \code | ||
20 | * void volk_32fc_deinterleave_real_32f(float* iBuffer, const lv_32fc_t* complexVector, | ||
21 | * unsigned int num_points) \endcode | ||
22 | * | ||
23 | * \b Inputs | ||
24 | * \li complexVector: The complex input vector. | ||
25 | * \li num_points: The number of complex data values to be deinterleaved. | ||
26 | * | ||
27 | * \b Outputs | ||
28 | * \li iBuffer: The I buffer output data. | ||
29 | * | ||
30 | * \b Example | ||
31 | * Generate complex numbers around the top half of the unit circle and | ||
32 | * extract all of the real parts to a float buffer. | ||
33 | * \code | ||
34 | * int N = 10; | ||
35 | * unsigned int alignment = volk_get_alignment(); | ||
36 | * lv_32fc_t* in = (lv_32fc_t*)volk_malloc(sizeof(lv_32fc_t)*N, alignment); | ||
37 | * float* re = (float*)volk_malloc(sizeof(float)*N, alignment); | ||
38 | * | ||
39 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
40 | * float real = 2.f * ((float)ii / (float)N) - 1.f; | ||
41 | * float imag = std::sqrt(1.f - real * real); | ||
42 | * in[ii] = lv_cmake(real, imag); | ||
43 | * } | ||
44 | * | ||
45 | * volk_32fc_deinterleave_real_32f(re, in, N); | ||
46 | * | ||
47 | * printf(" real part\n"); | ||
48 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
49 | * printf("out(%i) = %+.1f\n", ii, re[ii]); | ||
50 | * } | ||
51 | * | ||
52 | * volk_free(in); | ||
53 | * volk_free(re); | ||
54 | * \endcode | ||
55 | */ | ||
56 | |||
57 | #ifndef INCLUDED_volk_32fc_deinterleave_real_32f_a_H | ||
58 | #define INCLUDED_volk_32fc_deinterleave_real_32f_a_H | ||
59 | |||
60 | #include <inttypes.h> | ||
61 | #include <stdio.h> | ||
62 | |||
63 | #ifdef LV_HAVE_AVX2 | ||
64 | #include <immintrin.h> | ||
65 | |||
66 | 2 | static inline void volk_32fc_deinterleave_real_32f_a_avx2(float* iBuffer, | |
67 | const lv_32fc_t* complexVector, | ||
68 | unsigned int num_points) | ||
69 | { | ||
70 | 2 | unsigned int number = 0; | |
71 | 2 | const unsigned int eighthPoints = num_points / 8; | |
72 | |||
73 | 2 | const float* complexVectorPtr = (const float*)complexVector; | |
74 | 2 | float* iBufferPtr = iBuffer; | |
75 | |||
76 | __m256 cplxValue1, cplxValue2; | ||
77 | __m256 iValue; | ||
78 | 2 | __m256i idx = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0); | |
79 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
80 | |||
81 | 32766 | cplxValue1 = _mm256_load_ps(complexVectorPtr); | |
82 | 32766 | complexVectorPtr += 8; | |
83 | |||
84 | 32766 | cplxValue2 = _mm256_load_ps(complexVectorPtr); | |
85 | 32766 | complexVectorPtr += 8; | |
86 | |||
87 | // Arrange in i1i2i3i4 format | ||
88 | 32766 | iValue = _mm256_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(2, 0, 2, 0)); | |
89 | 32766 | iValue = _mm256_permutevar8x32_ps(iValue, idx); | |
90 | |||
91 | _mm256_store_ps(iBufferPtr, iValue); | ||
92 | |||
93 | 32766 | iBufferPtr += 8; | |
94 | } | ||
95 | |||
96 | 2 | number = eighthPoints * 8; | |
97 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
98 | 14 | *iBufferPtr++ = *complexVectorPtr++; | |
99 | 14 | complexVectorPtr++; | |
100 | } | ||
101 | 2 | } | |
102 | #endif /* LV_HAVE_AVX2 */ | ||
103 | |||
104 | #ifdef LV_HAVE_SSE | ||
105 | #include <xmmintrin.h> | ||
106 | |||
107 | 2 | static inline void volk_32fc_deinterleave_real_32f_a_sse(float* iBuffer, | |
108 | const lv_32fc_t* complexVector, | ||
109 | unsigned int num_points) | ||
110 | { | ||
111 | 2 | unsigned int number = 0; | |
112 | 2 | const unsigned int quarterPoints = num_points / 4; | |
113 | |||
114 | 2 | const float* complexVectorPtr = (const float*)complexVector; | |
115 | 2 | float* iBufferPtr = iBuffer; | |
116 | |||
117 | __m128 cplxValue1, cplxValue2, iValue; | ||
118 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
119 | |||
120 | 65534 | cplxValue1 = _mm_load_ps(complexVectorPtr); | |
121 | 65534 | complexVectorPtr += 4; | |
122 | |||
123 | 65534 | cplxValue2 = _mm_load_ps(complexVectorPtr); | |
124 | 65534 | complexVectorPtr += 4; | |
125 | |||
126 | // Arrange in i1i2i3i4 format | ||
127 | 65534 | iValue = _mm_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(2, 0, 2, 0)); | |
128 | |||
129 | _mm_store_ps(iBufferPtr, iValue); | ||
130 | |||
131 | 65534 | iBufferPtr += 4; | |
132 | } | ||
133 | |||
134 | 2 | number = quarterPoints * 4; | |
135 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
136 | 6 | *iBufferPtr++ = *complexVectorPtr++; | |
137 | 6 | complexVectorPtr++; | |
138 | } | ||
139 | 2 | } | |
140 | #endif /* LV_HAVE_SSE */ | ||
141 | |||
142 | |||
143 | #ifdef LV_HAVE_GENERIC | ||
144 | |||
145 | 2 | static inline void volk_32fc_deinterleave_real_32f_generic(float* iBuffer, | |
146 | const lv_32fc_t* complexVector, | ||
147 | unsigned int num_points) | ||
148 | { | ||
149 | 2 | unsigned int number = 0; | |
150 | 2 | const float* complexVectorPtr = (float*)complexVector; | |
151 | 2 | float* iBufferPtr = iBuffer; | |
152 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
153 | 262142 | *iBufferPtr++ = *complexVectorPtr++; | |
154 | 262142 | complexVectorPtr++; | |
155 | } | ||
156 | 2 | } | |
157 | #endif /* LV_HAVE_GENERIC */ | ||
158 | |||
159 | |||
160 | #ifdef LV_HAVE_NEON | ||
161 | #include <arm_neon.h> | ||
162 | |||
163 | static inline void volk_32fc_deinterleave_real_32f_neon(float* iBuffer, | ||
164 | const lv_32fc_t* complexVector, | ||
165 | unsigned int num_points) | ||
166 | { | ||
167 | unsigned int number = 0; | ||
168 | unsigned int quarter_points = num_points / 4; | ||
169 | const float* complexVectorPtr = (float*)complexVector; | ||
170 | float* iBufferPtr = iBuffer; | ||
171 | float32x4x2_t complexInput; | ||
172 | |||
173 | for (number = 0; number < quarter_points; number++) { | ||
174 | complexInput = vld2q_f32(complexVectorPtr); | ||
175 | vst1q_f32(iBufferPtr, complexInput.val[0]); | ||
176 | complexVectorPtr += 8; | ||
177 | iBufferPtr += 4; | ||
178 | } | ||
179 | |||
180 | for (number = quarter_points * 4; number < num_points; number++) { | ||
181 | *iBufferPtr++ = *complexVectorPtr++; | ||
182 | complexVectorPtr++; | ||
183 | } | ||
184 | } | ||
185 | #endif /* LV_HAVE_NEON */ | ||
186 | |||
187 | #endif /* INCLUDED_volk_32fc_deinterleave_real_32f_a_H */ | ||
188 | |||
189 | |||
190 | #ifndef INCLUDED_volk_32fc_deinterleave_real_32f_u_H | ||
191 | #define INCLUDED_volk_32fc_deinterleave_real_32f_u_H | ||
192 | |||
193 | #include <inttypes.h> | ||
194 | #include <stdio.h> | ||
195 | |||
196 | #ifdef LV_HAVE_AVX2 | ||
197 | #include <immintrin.h> | ||
198 | |||
199 | 2 | static inline void volk_32fc_deinterleave_real_32f_u_avx2(float* iBuffer, | |
200 | const lv_32fc_t* complexVector, | ||
201 | unsigned int num_points) | ||
202 | { | ||
203 | 2 | unsigned int number = 0; | |
204 | 2 | const unsigned int eighthPoints = num_points / 8; | |
205 | |||
206 | 2 | const float* complexVectorPtr = (const float*)complexVector; | |
207 | 2 | float* iBufferPtr = iBuffer; | |
208 | |||
209 | __m256 cplxValue1, cplxValue2; | ||
210 | __m256 iValue; | ||
211 | 2 | __m256i idx = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0); | |
212 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
213 | |||
214 | 32766 | cplxValue1 = _mm256_loadu_ps(complexVectorPtr); | |
215 | 32766 | complexVectorPtr += 8; | |
216 | |||
217 | 32766 | cplxValue2 = _mm256_loadu_ps(complexVectorPtr); | |
218 | 32766 | complexVectorPtr += 8; | |
219 | |||
220 | // Arrange in i1i2i3i4 format | ||
221 | 32766 | iValue = _mm256_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(2, 0, 2, 0)); | |
222 | 32766 | iValue = _mm256_permutevar8x32_ps(iValue, idx); | |
223 | |||
224 | _mm256_storeu_ps(iBufferPtr, iValue); | ||
225 | |||
226 | 32766 | iBufferPtr += 8; | |
227 | } | ||
228 | |||
229 | 2 | number = eighthPoints * 8; | |
230 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
231 | 14 | *iBufferPtr++ = *complexVectorPtr++; | |
232 | 14 | complexVectorPtr++; | |
233 | } | ||
234 | 2 | } | |
235 | #endif /* LV_HAVE_AVX2 */ | ||
236 | |||
237 | #endif /* INCLUDED_volk_32fc_deinterleave_real_32f_u_H */ | ||
238 |