| 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_32f_x2_interleave_32fc | ||
| 12 | * | ||
| 13 | * \b Overview | ||
| 14 | * | ||
| 15 | * Takes input vector iBuffer as the real (inphase) part and input | ||
| 16 | * vector qBuffer as the imag (quadrature) part and combines them into | ||
| 17 | * a complex output vector. | ||
| 18 | * | ||
| 19 | * c[i] = complex(a[i], b[i]) | ||
| 20 | * | ||
| 21 | * <b>Dispatcher Prototype</b> | ||
| 22 | * \code | ||
| 23 | * void volk_32f_x2_interleave_32fc(lv_32fc_t* complexVector, const float* iBuffer, const | ||
| 24 | * float* qBuffer, unsigned int num_points) \endcode | ||
| 25 | * | ||
| 26 | * \b Inputs | ||
| 27 | * \li iBuffer: Input vector of samples for the real part. | ||
| 28 | * \li qBuffer: Input vector of samples for the imaginary part. | ||
| 29 | * \li num_points: The number of values in both input vectors. | ||
| 30 | * | ||
| 31 | * \b Outputs | ||
| 32 | * \li complexVector: The output vector of complex numbers. | ||
| 33 | * | ||
| 34 | * \b Example | ||
| 35 | * Generate the top half of the unit circle with real points equally spaced. | ||
| 36 | * \code | ||
| 37 | * int N = 10; | ||
| 38 | * unsigned int alignment = volk_get_alignment(); | ||
| 39 | * float* imag = (float*)volk_malloc(sizeof(float)*N, alignment); | ||
| 40 | * float* real = (float*)volk_malloc(sizeof(float)*N, alignment); | ||
| 41 | * lv_32fc_t* out = (lv_32fc_t*)volk_malloc(sizeof(lv_32fc_t)*N, alignment); | ||
| 42 | * | ||
| 43 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 44 | * real[ii] = 2.f * ((float)ii / (float)N) - 1.f; | ||
| 45 | * imag[ii] = std::sqrt(1.f - real[ii] * real[ii]); | ||
| 46 | * } | ||
| 47 | * | ||
| 48 | * volk_32f_x2_interleave_32fc(out, imag, real, N); | ||
| 49 | * | ||
| 50 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 51 | * printf("out[%u] = %1.2f + %1.2fj\n", ii, std::real(out[ii]), std::imag(out[ii])); | ||
| 52 | * } | ||
| 53 | * | ||
| 54 | * volk_free(imag); | ||
| 55 | * volk_free(real); | ||
| 56 | * volk_free(out); | ||
| 57 | * \endcode | ||
| 58 | */ | ||
| 59 | |||
| 60 | #ifndef INCLUDED_volk_32f_x2_interleave_32fc_a_H | ||
| 61 | #define INCLUDED_volk_32f_x2_interleave_32fc_a_H | ||
| 62 | |||
| 63 | #include <inttypes.h> | ||
| 64 | #include <stdio.h> | ||
| 65 | |||
| 66 | #ifdef LV_HAVE_AVX | ||
| 67 | #include <immintrin.h> | ||
| 68 | |||
| 69 | 2 | static inline void volk_32f_x2_interleave_32fc_a_avx(lv_32fc_t* complexVector, | |
| 70 | const float* iBuffer, | ||
| 71 | const float* qBuffer, | ||
| 72 | unsigned int num_points) | ||
| 73 | { | ||
| 74 | 2 | unsigned int number = 0; | |
| 75 | 2 | float* complexVectorPtr = (float*)complexVector; | |
| 76 | 2 | const float* iBufferPtr = iBuffer; | |
| 77 | 2 | const float* qBufferPtr = qBuffer; | |
| 78 | |||
| 79 | 2 | const uint64_t eighthPoints = num_points / 8; | |
| 80 | |||
| 81 | __m256 iValue, qValue, cplxValue1, cplxValue2, cplxValue; | ||
| 82 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 83 | 32766 | iValue = _mm256_load_ps(iBufferPtr); | |
| 84 | 32766 | qValue = _mm256_load_ps(qBufferPtr); | |
| 85 | |||
| 86 | // Interleaves the lower two values in the i and q variables into one buffer | ||
| 87 | 32766 | cplxValue1 = _mm256_unpacklo_ps(iValue, qValue); | |
| 88 | // Interleaves the upper two values in the i and q variables into one buffer | ||
| 89 | 32766 | cplxValue2 = _mm256_unpackhi_ps(iValue, qValue); | |
| 90 | |||
| 91 | 32766 | cplxValue = _mm256_permute2f128_ps(cplxValue1, cplxValue2, 0x20); | |
| 92 | _mm256_store_ps(complexVectorPtr, cplxValue); | ||
| 93 | 32766 | complexVectorPtr += 8; | |
| 94 | |||
| 95 | 32766 | cplxValue = _mm256_permute2f128_ps(cplxValue1, cplxValue2, 0x31); | |
| 96 | _mm256_store_ps(complexVectorPtr, cplxValue); | ||
| 97 | 32766 | complexVectorPtr += 8; | |
| 98 | |||
| 99 | 32766 | iBufferPtr += 8; | |
| 100 | 32766 | qBufferPtr += 8; | |
| 101 | } | ||
| 102 | |||
| 103 | 2 | number = eighthPoints * 8; | |
| 104 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 105 | 14 | *complexVectorPtr++ = *iBufferPtr++; | |
| 106 | 14 | *complexVectorPtr++ = *qBufferPtr++; | |
| 107 | } | ||
| 108 | 2 | } | |
| 109 | |||
| 110 | #endif /* LV_HAV_AVX */ | ||
| 111 | |||
| 112 | #ifdef LV_HAVE_SSE | ||
| 113 | #include <xmmintrin.h> | ||
| 114 | |||
| 115 | 2 | static inline void volk_32f_x2_interleave_32fc_a_sse(lv_32fc_t* complexVector, | |
| 116 | const float* iBuffer, | ||
| 117 | const float* qBuffer, | ||
| 118 | unsigned int num_points) | ||
| 119 | { | ||
| 120 | 2 | unsigned int number = 0; | |
| 121 | 2 | float* complexVectorPtr = (float*)complexVector; | |
| 122 | 2 | const float* iBufferPtr = iBuffer; | |
| 123 | 2 | const float* qBufferPtr = qBuffer; | |
| 124 | |||
| 125 | 2 | const uint64_t quarterPoints = num_points / 4; | |
| 126 | |||
| 127 | __m128 iValue, qValue, cplxValue; | ||
| 128 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 129 | 65534 | iValue = _mm_load_ps(iBufferPtr); | |
| 130 | 65534 | qValue = _mm_load_ps(qBufferPtr); | |
| 131 | |||
| 132 | // Interleaves the lower two values in the i and q variables into one buffer | ||
| 133 | 65534 | cplxValue = _mm_unpacklo_ps(iValue, qValue); | |
| 134 | _mm_store_ps(complexVectorPtr, cplxValue); | ||
| 135 | 65534 | complexVectorPtr += 4; | |
| 136 | |||
| 137 | // Interleaves the upper two values in the i and q variables into one buffer | ||
| 138 | 65534 | cplxValue = _mm_unpackhi_ps(iValue, qValue); | |
| 139 | _mm_store_ps(complexVectorPtr, cplxValue); | ||
| 140 | 65534 | complexVectorPtr += 4; | |
| 141 | |||
| 142 | 65534 | iBufferPtr += 4; | |
| 143 | 65534 | qBufferPtr += 4; | |
| 144 | } | ||
| 145 | |||
| 146 | 2 | number = quarterPoints * 4; | |
| 147 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 148 | 6 | *complexVectorPtr++ = *iBufferPtr++; | |
| 149 | 6 | *complexVectorPtr++ = *qBufferPtr++; | |
| 150 | } | ||
| 151 | 2 | } | |
| 152 | #endif /* LV_HAVE_SSE */ | ||
| 153 | |||
| 154 | |||
| 155 | #ifdef LV_HAVE_NEON | ||
| 156 | #include <arm_neon.h> | ||
| 157 | |||
| 158 | static inline void volk_32f_x2_interleave_32fc_neon(lv_32fc_t* complexVector, | ||
| 159 | const float* iBuffer, | ||
| 160 | const float* qBuffer, | ||
| 161 | unsigned int num_points) | ||
| 162 | { | ||
| 163 | unsigned int quarter_points = num_points / 4; | ||
| 164 | unsigned int number; | ||
| 165 | float* complexVectorPtr = (float*)complexVector; | ||
| 166 | |||
| 167 | float32x4x2_t complex_vec; | ||
| 168 | for (number = 0; number < quarter_points; ++number) { | ||
| 169 | complex_vec.val[0] = vld1q_f32(iBuffer); | ||
| 170 | complex_vec.val[1] = vld1q_f32(qBuffer); | ||
| 171 | vst2q_f32(complexVectorPtr, complex_vec); | ||
| 172 | iBuffer += 4; | ||
| 173 | qBuffer += 4; | ||
| 174 | complexVectorPtr += 8; | ||
| 175 | } | ||
| 176 | |||
| 177 | for (number = quarter_points * 4; number < num_points; ++number) { | ||
| 178 | *complexVectorPtr++ = *iBuffer++; | ||
| 179 | *complexVectorPtr++ = *qBuffer++; | ||
| 180 | } | ||
| 181 | } | ||
| 182 | #endif /* LV_HAVE_NEON */ | ||
| 183 | |||
| 184 | |||
| 185 | #ifdef LV_HAVE_GENERIC | ||
| 186 | |||
| 187 | 2 | static inline void volk_32f_x2_interleave_32fc_generic(lv_32fc_t* complexVector, | |
| 188 | const float* iBuffer, | ||
| 189 | const float* qBuffer, | ||
| 190 | unsigned int num_points) | ||
| 191 | { | ||
| 192 | 2 | float* complexVectorPtr = (float*)complexVector; | |
| 193 | 2 | const float* iBufferPtr = iBuffer; | |
| 194 | 2 | const float* qBufferPtr = qBuffer; | |
| 195 | unsigned int number; | ||
| 196 | |||
| 197 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
| 198 | 262142 | *complexVectorPtr++ = *iBufferPtr++; | |
| 199 | 262142 | *complexVectorPtr++ = *qBufferPtr++; | |
| 200 | } | ||
| 201 | 2 | } | |
| 202 | #endif /* LV_HAVE_GENERIC */ | ||
| 203 | |||
| 204 | |||
| 205 | #endif /* INCLUDED_volk_32f_x2_interleave_32fc_a_H */ | ||
| 206 | |||
| 207 | #ifndef INCLUDED_volk_32f_x2_interleave_32fc_u_H | ||
| 208 | #define INCLUDED_volk_32f_x2_interleave_32fc_u_H | ||
| 209 | |||
| 210 | #include <inttypes.h> | ||
| 211 | #include <stdio.h> | ||
| 212 | |||
| 213 | #ifdef LV_HAVE_AVX | ||
| 214 | #include <immintrin.h> | ||
| 215 | |||
| 216 | 2 | static inline void volk_32f_x2_interleave_32fc_u_avx(lv_32fc_t* complexVector, | |
| 217 | const float* iBuffer, | ||
| 218 | const float* qBuffer, | ||
| 219 | unsigned int num_points) | ||
| 220 | { | ||
| 221 | 2 | unsigned int number = 0; | |
| 222 | 2 | float* complexVectorPtr = (float*)complexVector; | |
| 223 | 2 | const float* iBufferPtr = iBuffer; | |
| 224 | 2 | const float* qBufferPtr = qBuffer; | |
| 225 | |||
| 226 | 2 | const uint64_t eighthPoints = num_points / 8; | |
| 227 | |||
| 228 | __m256 iValue, qValue, cplxValue1, cplxValue2, cplxValue; | ||
| 229 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 230 | 32766 | iValue = _mm256_loadu_ps(iBufferPtr); | |
| 231 | 32766 | qValue = _mm256_loadu_ps(qBufferPtr); | |
| 232 | |||
| 233 | // Interleaves the lower two values in the i and q variables into one buffer | ||
| 234 | 32766 | cplxValue1 = _mm256_unpacklo_ps(iValue, qValue); | |
| 235 | // Interleaves the upper two values in the i and q variables into one buffer | ||
| 236 | 32766 | cplxValue2 = _mm256_unpackhi_ps(iValue, qValue); | |
| 237 | |||
| 238 | 32766 | cplxValue = _mm256_permute2f128_ps(cplxValue1, cplxValue2, 0x20); | |
| 239 | _mm256_storeu_ps(complexVectorPtr, cplxValue); | ||
| 240 | 32766 | complexVectorPtr += 8; | |
| 241 | |||
| 242 | 32766 | cplxValue = _mm256_permute2f128_ps(cplxValue1, cplxValue2, 0x31); | |
| 243 | _mm256_storeu_ps(complexVectorPtr, cplxValue); | ||
| 244 | 32766 | complexVectorPtr += 8; | |
| 245 | |||
| 246 | 32766 | iBufferPtr += 8; | |
| 247 | 32766 | qBufferPtr += 8; | |
| 248 | } | ||
| 249 | |||
| 250 | 2 | number = eighthPoints * 8; | |
| 251 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 252 | 14 | *complexVectorPtr++ = *iBufferPtr++; | |
| 253 | 14 | *complexVectorPtr++ = *qBufferPtr++; | |
| 254 | } | ||
| 255 | 2 | } | |
| 256 | #endif /* LV_HAVE_AVX */ | ||
| 257 | |||
| 258 | #endif /* INCLUDED_volk_32f_x2_interleave_32fc_u_H */ | ||
| 259 |