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_imag_32f | ||
12 | * | ||
13 | * \b Overview | ||
14 | * | ||
15 | * Deinterleaves the complex floating point vector and return the imaginary | ||
16 | * part (quadrature) of the samples. | ||
17 | * | ||
18 | * <b>Dispatcher Prototype</b> | ||
19 | * \code | ||
20 | * void volk_32fc_deinterleave_image_32f(float* qBuffer, 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 qBuffer: The Q 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 imaginary 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* im = (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_imag_32f(im, in, N); | ||
46 | * | ||
47 | * printf(" imaginary part\n"); | ||
48 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
49 | * printf("out(%i) = %+.1f\n", ii, im[ii]); | ||
50 | * } | ||
51 | * | ||
52 | * volk_free(in); | ||
53 | * volk_free(im); | ||
54 | * \endcode | ||
55 | */ | ||
56 | |||
57 | #ifndef INCLUDED_volk_32fc_deinterleave_imag_32f_a_H | ||
58 | #define INCLUDED_volk_32fc_deinterleave_imag_32f_a_H | ||
59 | |||
60 | #include <inttypes.h> | ||
61 | #include <stdio.h> | ||
62 | |||
63 | #ifdef LV_HAVE_AVX | ||
64 | #include <immintrin.h> | ||
65 | |||
66 | 2 | static inline void volk_32fc_deinterleave_imag_32f_a_avx(float* qBuffer, | |
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 | 2 | const float* complexVectorPtr = (const float*)complexVector; | |
73 | 2 | float* qBufferPtr = qBuffer; | |
74 | |||
75 | __m256 cplxValue1, cplxValue2, complex1, complex2, qValue; | ||
76 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
77 | |||
78 | 32766 | cplxValue1 = _mm256_load_ps(complexVectorPtr); | |
79 | 32766 | complexVectorPtr += 8; | |
80 | |||
81 | 32766 | cplxValue2 = _mm256_load_ps(complexVectorPtr); | |
82 | 32766 | complexVectorPtr += 8; | |
83 | |||
84 | 32766 | complex1 = _mm256_permute2f128_ps(cplxValue1, cplxValue2, 0x20); | |
85 | 32766 | complex2 = _mm256_permute2f128_ps(cplxValue1, cplxValue2, 0x31); | |
86 | |||
87 | // Arrange in q1q2q3q4 format | ||
88 | 32766 | qValue = _mm256_shuffle_ps(complex1, complex2, 0xdd); | |
89 | |||
90 | _mm256_store_ps(qBufferPtr, qValue); | ||
91 | |||
92 | 32766 | qBufferPtr += 8; | |
93 | } | ||
94 | |||
95 | 2 | number = eighthPoints * 8; | |
96 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
97 | 14 | complexVectorPtr++; | |
98 | 14 | *qBufferPtr++ = *complexVectorPtr++; | |
99 | } | ||
100 | 2 | } | |
101 | #endif /* LV_HAVE_AVX */ | ||
102 | |||
103 | #ifdef LV_HAVE_SSE | ||
104 | #include <xmmintrin.h> | ||
105 | |||
106 | 2 | static inline void volk_32fc_deinterleave_imag_32f_a_sse(float* qBuffer, | |
107 | const lv_32fc_t* complexVector, | ||
108 | unsigned int num_points) | ||
109 | { | ||
110 | 2 | unsigned int number = 0; | |
111 | 2 | const unsigned int quarterPoints = num_points / 4; | |
112 | |||
113 | 2 | const float* complexVectorPtr = (const float*)complexVector; | |
114 | 2 | float* qBufferPtr = qBuffer; | |
115 | |||
116 | __m128 cplxValue1, cplxValue2, iValue; | ||
117 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
118 | |||
119 | 65534 | cplxValue1 = _mm_load_ps(complexVectorPtr); | |
120 | 65534 | complexVectorPtr += 4; | |
121 | |||
122 | 65534 | cplxValue2 = _mm_load_ps(complexVectorPtr); | |
123 | 65534 | complexVectorPtr += 4; | |
124 | |||
125 | // Arrange in q1q2q3q4 format | ||
126 | 65534 | iValue = _mm_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(3, 1, 3, 1)); | |
127 | |||
128 | _mm_store_ps(qBufferPtr, iValue); | ||
129 | |||
130 | 65534 | qBufferPtr += 4; | |
131 | } | ||
132 | |||
133 | 2 | number = quarterPoints * 4; | |
134 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
135 | 6 | complexVectorPtr++; | |
136 | 6 | *qBufferPtr++ = *complexVectorPtr++; | |
137 | } | ||
138 | 2 | } | |
139 | #endif /* LV_HAVE_SSE */ | ||
140 | |||
141 | #ifdef LV_HAVE_NEON | ||
142 | #include <arm_neon.h> | ||
143 | |||
144 | static inline void volk_32fc_deinterleave_imag_32f_neon(float* qBuffer, | ||
145 | const lv_32fc_t* complexVector, | ||
146 | unsigned int num_points) | ||
147 | { | ||
148 | unsigned int number = 0; | ||
149 | unsigned int quarter_points = num_points / 4; | ||
150 | const float* complexVectorPtr = (float*)complexVector; | ||
151 | float* qBufferPtr = qBuffer; | ||
152 | float32x4x2_t complexInput; | ||
153 | |||
154 | for (number = 0; number < quarter_points; number++) { | ||
155 | complexInput = vld2q_f32(complexVectorPtr); | ||
156 | vst1q_f32(qBufferPtr, complexInput.val[1]); | ||
157 | complexVectorPtr += 8; | ||
158 | qBufferPtr += 4; | ||
159 | } | ||
160 | |||
161 | for (number = quarter_points * 4; number < num_points; number++) { | ||
162 | complexVectorPtr++; | ||
163 | *qBufferPtr++ = *complexVectorPtr++; | ||
164 | } | ||
165 | } | ||
166 | #endif /* LV_HAVE_NEON */ | ||
167 | |||
168 | #ifdef LV_HAVE_GENERIC | ||
169 | |||
170 | 2 | static inline void volk_32fc_deinterleave_imag_32f_generic(float* qBuffer, | |
171 | const lv_32fc_t* complexVector, | ||
172 | unsigned int num_points) | ||
173 | { | ||
174 | 2 | unsigned int number = 0; | |
175 | 2 | const float* complexVectorPtr = (float*)complexVector; | |
176 | 2 | float* qBufferPtr = qBuffer; | |
177 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
178 | 262142 | complexVectorPtr++; | |
179 | 262142 | *qBufferPtr++ = *complexVectorPtr++; | |
180 | } | ||
181 | 2 | } | |
182 | #endif /* LV_HAVE_GENERIC */ | ||
183 | |||
184 | |||
185 | #endif /* INCLUDED_volk_32fc_deinterleave_imag_32f_a_H */ | ||
186 | |||
187 | #ifndef INCLUDED_volk_32fc_deinterleave_imag_32f_u_H | ||
188 | #define INCLUDED_volk_32fc_deinterleave_imag_32f_u_H | ||
189 | |||
190 | #include <inttypes.h> | ||
191 | #include <stdio.h> | ||
192 | |||
193 | #ifdef LV_HAVE_AVX | ||
194 | #include <immintrin.h> | ||
195 | |||
196 | 2 | static inline void volk_32fc_deinterleave_imag_32f_u_avx(float* qBuffer, | |
197 | const lv_32fc_t* complexVector, | ||
198 | unsigned int num_points) | ||
199 | { | ||
200 | 2 | unsigned int number = 0; | |
201 | 2 | const unsigned int eighthPoints = num_points / 8; | |
202 | 2 | const float* complexVectorPtr = (const float*)complexVector; | |
203 | 2 | float* qBufferPtr = qBuffer; | |
204 | |||
205 | __m256 cplxValue1, cplxValue2, complex1, complex2, qValue; | ||
206 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
207 | |||
208 | 32766 | cplxValue1 = _mm256_loadu_ps(complexVectorPtr); | |
209 | 32766 | complexVectorPtr += 8; | |
210 | |||
211 | 32766 | cplxValue2 = _mm256_loadu_ps(complexVectorPtr); | |
212 | 32766 | complexVectorPtr += 8; | |
213 | |||
214 | 32766 | complex1 = _mm256_permute2f128_ps(cplxValue1, cplxValue2, 0x20); | |
215 | 32766 | complex2 = _mm256_permute2f128_ps(cplxValue1, cplxValue2, 0x31); | |
216 | |||
217 | // Arrange in q1q2q3q4 format | ||
218 | 32766 | qValue = _mm256_shuffle_ps(complex1, complex2, 0xdd); | |
219 | |||
220 | _mm256_storeu_ps(qBufferPtr, qValue); | ||
221 | |||
222 | 32766 | qBufferPtr += 8; | |
223 | } | ||
224 | |||
225 | 2 | number = eighthPoints * 8; | |
226 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
227 | 14 | complexVectorPtr++; | |
228 | 14 | *qBufferPtr++ = *complexVectorPtr++; | |
229 | } | ||
230 | 2 | } | |
231 | #endif /* LV_HAVE_AVX */ | ||
232 | #endif /* INCLUDED_volk_32fc_deinterleave_imag_32f_u_H */ | ||
233 |