| 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_convert_64f | ||
| 12 | * | ||
| 13 | * \b Overview | ||
| 14 | * | ||
| 15 | * Converts float values into doubles. | ||
| 16 | * | ||
| 17 | * <b>Dispatcher Prototype</b> | ||
| 18 | * \code | ||
| 19 | * void volk_32f_convert_64f(double* outputVector, const float* inputVector, unsigned int | ||
| 20 | * num_points) \endcode | ||
| 21 | * | ||
| 22 | * \b Inputs | ||
| 23 | * \li inputVector: The vector of floats to convert to doubles. | ||
| 24 | * \li num_points: The number of data points. | ||
| 25 | * | ||
| 26 | * \b Outputs | ||
| 27 | * \li outputVector: returns the converted doubles. | ||
| 28 | * | ||
| 29 | * \b Example | ||
| 30 | * Generate floats and convert them to doubles. | ||
| 31 | * \code | ||
| 32 | * int N = 10; | ||
| 33 | * unsigned int alignment = volk_get_alignment(); | ||
| 34 | * float* in = (float*)volk_malloc(sizeof(float)*N, alignment); | ||
| 35 | * double* out = (double*)volk_malloc(sizeof(double)*N, alignment); | ||
| 36 | * | ||
| 37 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 38 | * in[ii] = (float)ii; | ||
| 39 | * } | ||
| 40 | * | ||
| 41 | * volk_32f_convert_64f(out, in, N); | ||
| 42 | * | ||
| 43 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 44 | * printf("out(%i) = %g\n", ii, out[ii]); | ||
| 45 | * } | ||
| 46 | * | ||
| 47 | * volk_free(in); | ||
| 48 | * volk_free(out); | ||
| 49 | * \endcode | ||
| 50 | */ | ||
| 51 | |||
| 52 | |||
| 53 | #ifndef INCLUDED_volk_32f_convert_64f_u_H | ||
| 54 | #define INCLUDED_volk_32f_convert_64f_u_H | ||
| 55 | |||
| 56 | #include <inttypes.h> | ||
| 57 | #include <stdio.h> | ||
| 58 | |||
| 59 | #ifdef LV_HAVE_AVX | ||
| 60 | #include <immintrin.h> | ||
| 61 | |||
| 62 | 2 | static inline void volk_32f_convert_64f_u_avx(double* outputVector, | |
| 63 | const float* inputVector, | ||
| 64 | unsigned int num_points) | ||
| 65 | { | ||
| 66 | 2 | unsigned int number = 0; | |
| 67 | |||
| 68 | 2 | const unsigned int quarterPoints = num_points / 4; | |
| 69 | |||
| 70 | 2 | const float* inputVectorPtr = (const float*)inputVector; | |
| 71 | 2 | double* outputVectorPtr = outputVector; | |
| 72 | __m256d ret; | ||
| 73 | __m128 inputVal; | ||
| 74 | |||
| 75 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 76 | 65534 | inputVal = _mm_loadu_ps(inputVectorPtr); | |
| 77 | 65534 | inputVectorPtr += 4; | |
| 78 | |||
| 79 | 65534 | ret = _mm256_cvtps_pd(inputVal); | |
| 80 | _mm256_storeu_pd(outputVectorPtr, ret); | ||
| 81 | |||
| 82 | 65534 | outputVectorPtr += 4; | |
| 83 | } | ||
| 84 | |||
| 85 | 2 | number = quarterPoints * 4; | |
| 86 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 87 | 6 | outputVector[number] = (double)(inputVector[number]); | |
| 88 | } | ||
| 89 | 2 | } | |
| 90 | |||
| 91 | #endif /* LV_HAVE_AVX */ | ||
| 92 | |||
| 93 | #ifdef LV_HAVE_SSE2 | ||
| 94 | #include <emmintrin.h> | ||
| 95 | |||
| 96 | 2 | static inline void volk_32f_convert_64f_u_sse2(double* outputVector, | |
| 97 | const float* inputVector, | ||
| 98 | unsigned int num_points) | ||
| 99 | { | ||
| 100 | 2 | unsigned int number = 0; | |
| 101 | |||
| 102 | 2 | const unsigned int quarterPoints = num_points / 4; | |
| 103 | |||
| 104 | 2 | const float* inputVectorPtr = (const float*)inputVector; | |
| 105 | 2 | double* outputVectorPtr = outputVector; | |
| 106 | __m128d ret; | ||
| 107 | __m128 inputVal; | ||
| 108 | |||
| 109 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 110 | 65534 | inputVal = _mm_loadu_ps(inputVectorPtr); | |
| 111 | 65534 | inputVectorPtr += 4; | |
| 112 | |||
| 113 | 65534 | ret = _mm_cvtps_pd(inputVal); | |
| 114 | |||
| 115 | _mm_storeu_pd(outputVectorPtr, ret); | ||
| 116 | 65534 | outputVectorPtr += 2; | |
| 117 | |||
| 118 | 65534 | inputVal = _mm_movehl_ps(inputVal, inputVal); | |
| 119 | |||
| 120 | 65534 | ret = _mm_cvtps_pd(inputVal); | |
| 121 | |||
| 122 | _mm_storeu_pd(outputVectorPtr, ret); | ||
| 123 | 65534 | outputVectorPtr += 2; | |
| 124 | } | ||
| 125 | |||
| 126 | 2 | number = quarterPoints * 4; | |
| 127 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 128 | 6 | outputVector[number] = (double)(inputVector[number]); | |
| 129 | } | ||
| 130 | 2 | } | |
| 131 | #endif /* LV_HAVE_SSE2 */ | ||
| 132 | |||
| 133 | |||
| 134 | #ifdef LV_HAVE_GENERIC | ||
| 135 | |||
| 136 | 2 | static inline void volk_32f_convert_64f_generic(double* outputVector, | |
| 137 | const float* inputVector, | ||
| 138 | unsigned int num_points) | ||
| 139 | { | ||
| 140 | 2 | double* outputVectorPtr = outputVector; | |
| 141 | 2 | const float* inputVectorPtr = inputVector; | |
| 142 | 2 | unsigned int number = 0; | |
| 143 | |||
| 144 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
| 145 | 262142 | *outputVectorPtr++ = ((double)(*inputVectorPtr++)); | |
| 146 | } | ||
| 147 | 2 | } | |
| 148 | #endif /* LV_HAVE_GENERIC */ | ||
| 149 | |||
| 150 | |||
| 151 | #endif /* INCLUDED_volk_32f_convert_64f_u_H */ | ||
| 152 | |||
| 153 | |||
| 154 | #ifndef INCLUDED_volk_32f_convert_64f_a_H | ||
| 155 | #define INCLUDED_volk_32f_convert_64f_a_H | ||
| 156 | |||
| 157 | #include <inttypes.h> | ||
| 158 | #include <stdio.h> | ||
| 159 | |||
| 160 | #ifdef LV_HAVE_AVX | ||
| 161 | #include <immintrin.h> | ||
| 162 | |||
| 163 | 2 | static inline void volk_32f_convert_64f_a_avx(double* outputVector, | |
| 164 | const float* inputVector, | ||
| 165 | unsigned int num_points) | ||
| 166 | { | ||
| 167 | 2 | unsigned int number = 0; | |
| 168 | |||
| 169 | 2 | const unsigned int quarterPoints = num_points / 4; | |
| 170 | |||
| 171 | 2 | const float* inputVectorPtr = (const float*)inputVector; | |
| 172 | 2 | double* outputVectorPtr = outputVector; | |
| 173 | __m256d ret; | ||
| 174 | __m128 inputVal; | ||
| 175 | |||
| 176 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 177 | 65534 | inputVal = _mm_load_ps(inputVectorPtr); | |
| 178 | 65534 | inputVectorPtr += 4; | |
| 179 | |||
| 180 | 65534 | ret = _mm256_cvtps_pd(inputVal); | |
| 181 | _mm256_store_pd(outputVectorPtr, ret); | ||
| 182 | |||
| 183 | 65534 | outputVectorPtr += 4; | |
| 184 | } | ||
| 185 | |||
| 186 | 2 | number = quarterPoints * 4; | |
| 187 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 188 | 6 | outputVector[number] = (double)(inputVector[number]); | |
| 189 | } | ||
| 190 | 2 | } | |
| 191 | #endif /* LV_HAVE_AVX */ | ||
| 192 | |||
| 193 | #ifdef LV_HAVE_SSE2 | ||
| 194 | #include <emmintrin.h> | ||
| 195 | |||
| 196 | 2 | static inline void volk_32f_convert_64f_a_sse2(double* outputVector, | |
| 197 | const float* inputVector, | ||
| 198 | unsigned int num_points) | ||
| 199 | { | ||
| 200 | 2 | unsigned int number = 0; | |
| 201 | |||
| 202 | 2 | const unsigned int quarterPoints = num_points / 4; | |
| 203 | |||
| 204 | 2 | const float* inputVectorPtr = (const float*)inputVector; | |
| 205 | 2 | double* outputVectorPtr = outputVector; | |
| 206 | __m128d ret; | ||
| 207 | __m128 inputVal; | ||
| 208 | |||
| 209 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 210 | 65534 | inputVal = _mm_load_ps(inputVectorPtr); | |
| 211 | 65534 | inputVectorPtr += 4; | |
| 212 | |||
| 213 | 65534 | ret = _mm_cvtps_pd(inputVal); | |
| 214 | |||
| 215 | _mm_store_pd(outputVectorPtr, ret); | ||
| 216 | 65534 | outputVectorPtr += 2; | |
| 217 | |||
| 218 | 65534 | inputVal = _mm_movehl_ps(inputVal, inputVal); | |
| 219 | |||
| 220 | 65534 | ret = _mm_cvtps_pd(inputVal); | |
| 221 | |||
| 222 | _mm_store_pd(outputVectorPtr, ret); | ||
| 223 | 65534 | outputVectorPtr += 2; | |
| 224 | } | ||
| 225 | |||
| 226 | 2 | number = quarterPoints * 4; | |
| 227 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 228 | 6 | outputVector[number] = (double)(inputVector[number]); | |
| 229 | } | ||
| 230 | 2 | } | |
| 231 | #endif /* LV_HAVE_SSE2 */ | ||
| 232 | |||
| 233 | |||
| 234 | #ifdef LV_HAVE_GENERIC | ||
| 235 | |||
| 236 | 2 | static inline void volk_32f_convert_64f_a_generic(double* outputVector, | |
| 237 | const float* inputVector, | ||
| 238 | unsigned int num_points) | ||
| 239 | { | ||
| 240 | 2 | double* outputVectorPtr = outputVector; | |
| 241 | 2 | const float* inputVectorPtr = inputVector; | |
| 242 | 2 | unsigned int number = 0; | |
| 243 | |||
| 244 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
| 245 | 262142 | *outputVectorPtr++ = ((double)(*inputVectorPtr++)); | |
| 246 | } | ||
| 247 | 2 | } | |
| 248 | #endif /* LV_HAVE_GENERIC */ | ||
| 249 | |||
| 250 | |||
| 251 | #endif /* INCLUDED_volk_32f_convert_64f_a_H */ | ||
| 252 |