| 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_32f_multiply_32fc | ||
| 12 | * | ||
| 13 | * \b Overview | ||
| 14 | * | ||
| 15 | * Multiplies a complex vector by a floating point vector and returns | ||
| 16 | * the complex result. | ||
| 17 | * | ||
| 18 | * <b>Dispatcher Prototype</b> | ||
| 19 | * \code | ||
| 20 | * void volk_32fc_32f_multiply_32fc(lv_32fc_t* cVector, const lv_32fc_t* aVector, const | ||
| 21 | * float* bVector, unsigned int num_points); \endcode | ||
| 22 | * | ||
| 23 | * \b Inputs | ||
| 24 | * \li aVector: The input vector of complex floats. | ||
| 25 | * \li bVector: The input vector of floats. | ||
| 26 | * \li num_points: The number of data points. | ||
| 27 | * | ||
| 28 | * \b Outputs | ||
| 29 | * \li outputVector: The output vector complex floats. | ||
| 30 | * | ||
| 31 | * \b Example | ||
| 32 | * \code | ||
| 33 | * int N = 10000; | ||
| 34 | * | ||
| 35 | * volk_32fc_32f_multiply_32fc(); | ||
| 36 | * | ||
| 37 | * volk_free(x); | ||
| 38 | * volk_free(t); | ||
| 39 | * \endcode | ||
| 40 | */ | ||
| 41 | |||
| 42 | #ifndef INCLUDED_volk_32fc_32f_multiply_32fc_a_H | ||
| 43 | #define INCLUDED_volk_32fc_32f_multiply_32fc_a_H | ||
| 44 | |||
| 45 | #include <inttypes.h> | ||
| 46 | #include <stdio.h> | ||
| 47 | |||
| 48 | #ifdef LV_HAVE_AVX | ||
| 49 | #include <immintrin.h> | ||
| 50 | |||
| 51 | 2 | static inline void volk_32fc_32f_multiply_32fc_a_avx(lv_32fc_t* cVector, | |
| 52 | const lv_32fc_t* aVector, | ||
| 53 | const float* bVector, | ||
| 54 | unsigned int num_points) | ||
| 55 | { | ||
| 56 | 2 | unsigned int number = 0; | |
| 57 | 2 | const unsigned int eighthPoints = num_points / 8; | |
| 58 | |||
| 59 | 2 | lv_32fc_t* cPtr = cVector; | |
| 60 | 2 | const lv_32fc_t* aPtr = aVector; | |
| 61 | 2 | const float* bPtr = bVector; | |
| 62 | |||
| 63 | __m256 aVal1, aVal2, bVal, bVal1, bVal2, cVal1, cVal2; | ||
| 64 | |||
| 65 | 2 | __m256i permute_mask = _mm256_set_epi32(3, 3, 2, 2, 1, 1, 0, 0); | |
| 66 | |||
| 67 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 68 | |||
| 69 | 32766 | aVal1 = _mm256_load_ps((float*)aPtr); | |
| 70 | 32766 | aPtr += 4; | |
| 71 | |||
| 72 | 32766 | aVal2 = _mm256_load_ps((float*)aPtr); | |
| 73 | 32766 | aPtr += 4; | |
| 74 | |||
| 75 | 32766 | bVal = _mm256_load_ps(bPtr); // b0|b1|b2|b3|b4|b5|b6|b7 | |
| 76 | 32766 | bPtr += 8; | |
| 77 | |||
| 78 | 32766 | bVal1 = _mm256_permute2f128_ps(bVal, bVal, 0x00); // b0|b1|b2|b3|b0|b1|b2|b3 | |
| 79 | 32766 | bVal2 = _mm256_permute2f128_ps(bVal, bVal, 0x11); // b4|b5|b6|b7|b4|b5|b6|b7 | |
| 80 | |||
| 81 | 32766 | bVal1 = _mm256_permutevar_ps(bVal1, permute_mask); // b0|b0|b1|b1|b2|b2|b3|b3 | |
| 82 | 32766 | bVal2 = _mm256_permutevar_ps(bVal2, permute_mask); // b4|b4|b5|b5|b6|b6|b7|b7 | |
| 83 | |||
| 84 | 32766 | cVal1 = _mm256_mul_ps(aVal1, bVal1); | |
| 85 | 32766 | cVal2 = _mm256_mul_ps(aVal2, bVal2); | |
| 86 | |||
| 87 | _mm256_store_ps((float*)cPtr, | ||
| 88 | cVal1); // Store the results back into the C container | ||
| 89 | 32766 | cPtr += 4; | |
| 90 | |||
| 91 | _mm256_store_ps((float*)cPtr, | ||
| 92 | cVal2); // Store the results back into the C container | ||
| 93 | 32766 | cPtr += 4; | |
| 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 | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 99 | } | ||
| 100 | 2 | } | |
| 101 | #endif /* LV_HAVE_AVX */ | ||
| 102 | |||
| 103 | |||
| 104 | #ifdef LV_HAVE_SSE | ||
| 105 | #include <xmmintrin.h> | ||
| 106 | |||
| 107 | 2 | static inline void volk_32fc_32f_multiply_32fc_a_sse(lv_32fc_t* cVector, | |
| 108 | const lv_32fc_t* aVector, | ||
| 109 | const float* bVector, | ||
| 110 | unsigned int num_points) | ||
| 111 | { | ||
| 112 | 2 | unsigned int number = 0; | |
| 113 | 2 | const unsigned int quarterPoints = num_points / 4; | |
| 114 | |||
| 115 | 2 | lv_32fc_t* cPtr = cVector; | |
| 116 | 2 | const lv_32fc_t* aPtr = aVector; | |
| 117 | 2 | const float* bPtr = bVector; | |
| 118 | |||
| 119 | __m128 aVal1, aVal2, bVal, bVal1, bVal2, cVal; | ||
| 120 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 121 | |||
| 122 | 65534 | aVal1 = _mm_load_ps((const float*)aPtr); | |
| 123 | 65534 | aPtr += 2; | |
| 124 | |||
| 125 | 65534 | aVal2 = _mm_load_ps((const float*)aPtr); | |
| 126 | 65534 | aPtr += 2; | |
| 127 | |||
| 128 | 65534 | bVal = _mm_load_ps(bPtr); | |
| 129 | 65534 | bPtr += 4; | |
| 130 | |||
| 131 | 65534 | bVal1 = _mm_shuffle_ps(bVal, bVal, _MM_SHUFFLE(1, 1, 0, 0)); | |
| 132 | 65534 | bVal2 = _mm_shuffle_ps(bVal, bVal, _MM_SHUFFLE(3, 3, 2, 2)); | |
| 133 | |||
| 134 | 65534 | cVal = _mm_mul_ps(aVal1, bVal1); | |
| 135 | |||
| 136 | _mm_store_ps((float*)cPtr, cVal); // Store the results back into the C container | ||
| 137 | 65534 | cPtr += 2; | |
| 138 | |||
| 139 | 65534 | cVal = _mm_mul_ps(aVal2, bVal2); | |
| 140 | |||
| 141 | _mm_store_ps((float*)cPtr, cVal); // Store the results back into the C container | ||
| 142 | |||
| 143 | 65534 | cPtr += 2; | |
| 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 | *cPtr++ = (*aPtr++) * (*bPtr); | |
| 149 | 6 | bPtr++; | |
| 150 | } | ||
| 151 | 2 | } | |
| 152 | #endif /* LV_HAVE_SSE */ | ||
| 153 | |||
| 154 | |||
| 155 | #ifdef LV_HAVE_GENERIC | ||
| 156 | |||
| 157 | 2 | static inline void volk_32fc_32f_multiply_32fc_generic(lv_32fc_t* cVector, | |
| 158 | const lv_32fc_t* aVector, | ||
| 159 | const float* bVector, | ||
| 160 | unsigned int num_points) | ||
| 161 | { | ||
| 162 | 2 | lv_32fc_t* cPtr = cVector; | |
| 163 | 2 | const lv_32fc_t* aPtr = aVector; | |
| 164 | 2 | const float* bPtr = bVector; | |
| 165 | 2 | unsigned int number = 0; | |
| 166 | |||
| 167 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
| 168 | 262142 | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 169 | } | ||
| 170 | 2 | } | |
| 171 | #endif /* LV_HAVE_GENERIC */ | ||
| 172 | |||
| 173 | |||
| 174 | #ifdef LV_HAVE_NEON | ||
| 175 | #include <arm_neon.h> | ||
| 176 | |||
| 177 | static inline void volk_32fc_32f_multiply_32fc_neon(lv_32fc_t* cVector, | ||
| 178 | const lv_32fc_t* aVector, | ||
| 179 | const float* bVector, | ||
| 180 | unsigned int num_points) | ||
| 181 | { | ||
| 182 | lv_32fc_t* cPtr = cVector; | ||
| 183 | const lv_32fc_t* aPtr = aVector; | ||
| 184 | const float* bPtr = bVector; | ||
| 185 | unsigned int number = 0; | ||
| 186 | unsigned int quarter_points = num_points / 4; | ||
| 187 | |||
| 188 | float32x4x2_t inputVector, outputVector; | ||
| 189 | float32x4_t tapsVector; | ||
| 190 | for (number = 0; number < quarter_points; number++) { | ||
| 191 | inputVector = vld2q_f32((float*)aPtr); | ||
| 192 | tapsVector = vld1q_f32(bPtr); | ||
| 193 | |||
| 194 | outputVector.val[0] = vmulq_f32(inputVector.val[0], tapsVector); | ||
| 195 | outputVector.val[1] = vmulq_f32(inputVector.val[1], tapsVector); | ||
| 196 | |||
| 197 | vst2q_f32((float*)cPtr, outputVector); | ||
| 198 | aPtr += 4; | ||
| 199 | bPtr += 4; | ||
| 200 | cPtr += 4; | ||
| 201 | } | ||
| 202 | |||
| 203 | for (number = quarter_points * 4; number < num_points; number++) { | ||
| 204 | *cPtr++ = (*aPtr++) * (*bPtr++); | ||
| 205 | } | ||
| 206 | } | ||
| 207 | #endif /* LV_HAVE_NEON */ | ||
| 208 | |||
| 209 | |||
| 210 | #ifdef LV_HAVE_ORC | ||
| 211 | |||
| 212 | extern void volk_32fc_32f_multiply_32fc_a_orc_impl(lv_32fc_t* cVector, | ||
| 213 | const lv_32fc_t* aVector, | ||
| 214 | const float* bVector, | ||
| 215 | unsigned int num_points); | ||
| 216 | |||
| 217 | 2 | static inline void volk_32fc_32f_multiply_32fc_u_orc(lv_32fc_t* cVector, | |
| 218 | const lv_32fc_t* aVector, | ||
| 219 | const float* bVector, | ||
| 220 | unsigned int num_points) | ||
| 221 | { | ||
| 222 | 2 | volk_32fc_32f_multiply_32fc_a_orc_impl(cVector, aVector, bVector, num_points); | |
| 223 | 2 | } | |
| 224 | |||
| 225 | #endif /* LV_HAVE_GENERIC */ | ||
| 226 | |||
| 227 | |||
| 228 | #endif /* INCLUDED_volk_32fc_32f_multiply_32fc_a_H */ | ||
| 229 |