| 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_64f_convert_32f | ||
| 12 | * | ||
| 13 | * \b Overview | ||
| 14 | * | ||
| 15 | * Converts doubles into floats. | ||
| 16 | * | ||
| 17 | * <b>Dispatcher Prototype</b> | ||
| 18 | * \code | ||
| 19 | * void volk_64f_convert_32f(float* outputVector, const double* inputVector, unsigned int | ||
| 20 | * num_points) \endcode | ||
| 21 | * | ||
| 22 | * \b Inputs | ||
| 23 | * \li inputVector: The vector of doubles to convert to floats. | ||
| 24 | * \li num_points: The number of data points. | ||
| 25 | * | ||
| 26 | * \b Outputs | ||
| 27 | * \li outputVector: returns the converted floats. | ||
| 28 | * | ||
| 29 | * \b Example | ||
| 30 | * \code | ||
| 31 | * int N = 10; | ||
| 32 | * unsigned int alignment = volk_get_alignment(); | ||
| 33 | * double* increasing = (double*)volk_malloc(sizeof(double)*N, alignment); | ||
| 34 | * float* out = (float*)volk_malloc(sizeof(float)*N, alignment); | ||
| 35 | * | ||
| 36 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 37 | * increasing[ii] = (double)ii; | ||
| 38 | * } | ||
| 39 | * | ||
| 40 | * volk_64f_convert_32f(out, increasing, N); | ||
| 41 | * | ||
| 42 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 43 | * printf("out[%u] = %1.2f\n", ii, out[ii]); | ||
| 44 | * } | ||
| 45 | * | ||
| 46 | * volk_free(increasing); | ||
| 47 | * volk_free(out); | ||
| 48 | * \endcode | ||
| 49 | */ | ||
| 50 | |||
| 51 | #ifndef INCLUDED_volk_64f_convert_32f_u_H | ||
| 52 | #define INCLUDED_volk_64f_convert_32f_u_H | ||
| 53 | |||
| 54 | #include <inttypes.h> | ||
| 55 | #include <stdio.h> | ||
| 56 | |||
| 57 | #ifdef LV_HAVE_AVX512F | ||
| 58 | #include <immintrin.h> | ||
| 59 | |||
| 60 | ✗ | static inline void volk_64f_convert_32f_u_avx512f(float* outputVector, | |
| 61 | const double* inputVector, | ||
| 62 | unsigned int num_points) | ||
| 63 | { | ||
| 64 | ✗ | unsigned int number = 0; | |
| 65 | |||
| 66 | ✗ | const unsigned int oneSixteenthPoints = num_points / 16; | |
| 67 | |||
| 68 | ✗ | const double* inputVectorPtr = (const double*)inputVector; | |
| 69 | ✗ | float* outputVectorPtr = outputVector; | |
| 70 | __m256 ret1, ret2; | ||
| 71 | __m512d inputVal1, inputVal2; | ||
| 72 | |||
| 73 | ✗ | for (; number < oneSixteenthPoints; number++) { | |
| 74 | ✗ | inputVal1 = _mm512_loadu_pd(inputVectorPtr); | |
| 75 | ✗ | inputVectorPtr += 8; | |
| 76 | ✗ | inputVal2 = _mm512_loadu_pd(inputVectorPtr); | |
| 77 | ✗ | inputVectorPtr += 8; | |
| 78 | |||
| 79 | ✗ | ret1 = _mm512_cvtpd_ps(inputVal1); | |
| 80 | ✗ | ret2 = _mm512_cvtpd_ps(inputVal2); | |
| 81 | |||
| 82 | _mm256_storeu_ps(outputVectorPtr, ret1); | ||
| 83 | ✗ | outputVectorPtr += 8; | |
| 84 | |||
| 85 | _mm256_storeu_ps(outputVectorPtr, ret2); | ||
| 86 | ✗ | outputVectorPtr += 8; | |
| 87 | } | ||
| 88 | |||
| 89 | ✗ | number = oneSixteenthPoints * 16; | |
| 90 | ✗ | for (; number < num_points; number++) { | |
| 91 | ✗ | outputVector[number] = (float)(inputVector[number]); | |
| 92 | } | ||
| 93 | ✗ | } | |
| 94 | #endif /* LV_HAVE_AVX512F */ | ||
| 95 | |||
| 96 | |||
| 97 | #ifdef LV_HAVE_AVX | ||
| 98 | #include <immintrin.h> | ||
| 99 | |||
| 100 | 2 | static inline void volk_64f_convert_32f_u_avx(float* outputVector, | |
| 101 | const double* inputVector, | ||
| 102 | unsigned int num_points) | ||
| 103 | { | ||
| 104 | 2 | unsigned int number = 0; | |
| 105 | |||
| 106 | 2 | const unsigned int oneEightPoints = num_points / 8; | |
| 107 | |||
| 108 | 2 | const double* inputVectorPtr = (const double*)inputVector; | |
| 109 | 2 | float* outputVectorPtr = outputVector; | |
| 110 | __m128 ret1, ret2; | ||
| 111 | __m256d inputVal1, inputVal2; | ||
| 112 | |||
| 113 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < oneEightPoints; number++) { |
| 114 | 32766 | inputVal1 = _mm256_loadu_pd(inputVectorPtr); | |
| 115 | 32766 | inputVectorPtr += 4; | |
| 116 | 32766 | inputVal2 = _mm256_loadu_pd(inputVectorPtr); | |
| 117 | 32766 | inputVectorPtr += 4; | |
| 118 | |||
| 119 | 32766 | ret1 = _mm256_cvtpd_ps(inputVal1); | |
| 120 | 32766 | ret2 = _mm256_cvtpd_ps(inputVal2); | |
| 121 | |||
| 122 | _mm_storeu_ps(outputVectorPtr, ret1); | ||
| 123 | 32766 | outputVectorPtr += 4; | |
| 124 | |||
| 125 | _mm_storeu_ps(outputVectorPtr, ret2); | ||
| 126 | 32766 | outputVectorPtr += 4; | |
| 127 | } | ||
| 128 | |||
| 129 | 2 | number = oneEightPoints * 8; | |
| 130 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 131 | 14 | outputVector[number] = (float)(inputVector[number]); | |
| 132 | } | ||
| 133 | 2 | } | |
| 134 | #endif /* LV_HAVE_AVX */ | ||
| 135 | |||
| 136 | |||
| 137 | #ifdef LV_HAVE_SSE2 | ||
| 138 | #include <emmintrin.h> | ||
| 139 | |||
| 140 | 2 | static inline void volk_64f_convert_32f_u_sse2(float* outputVector, | |
| 141 | const double* inputVector, | ||
| 142 | unsigned int num_points) | ||
| 143 | { | ||
| 144 | 2 | unsigned int number = 0; | |
| 145 | |||
| 146 | 2 | const unsigned int quarterPoints = num_points / 4; | |
| 147 | |||
| 148 | 2 | const double* inputVectorPtr = (const double*)inputVector; | |
| 149 | 2 | float* outputVectorPtr = outputVector; | |
| 150 | __m128 ret, ret2; | ||
| 151 | __m128d inputVal1, inputVal2; | ||
| 152 | |||
| 153 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 154 | 65534 | inputVal1 = _mm_loadu_pd(inputVectorPtr); | |
| 155 | 65534 | inputVectorPtr += 2; | |
| 156 | 65534 | inputVal2 = _mm_loadu_pd(inputVectorPtr); | |
| 157 | 65534 | inputVectorPtr += 2; | |
| 158 | |||
| 159 | 65534 | ret = _mm_cvtpd_ps(inputVal1); | |
| 160 | 65534 | ret2 = _mm_cvtpd_ps(inputVal2); | |
| 161 | |||
| 162 | 65534 | ret = _mm_movelh_ps(ret, ret2); | |
| 163 | |||
| 164 | _mm_storeu_ps(outputVectorPtr, ret); | ||
| 165 | 65534 | outputVectorPtr += 4; | |
| 166 | } | ||
| 167 | |||
| 168 | 2 | number = quarterPoints * 4; | |
| 169 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 170 | 6 | outputVector[number] = (float)(inputVector[number]); | |
| 171 | } | ||
| 172 | 2 | } | |
| 173 | #endif /* LV_HAVE_SSE2 */ | ||
| 174 | |||
| 175 | |||
| 176 | #ifdef LV_HAVE_GENERIC | ||
| 177 | |||
| 178 | 2 | static inline void volk_64f_convert_32f_generic(float* outputVector, | |
| 179 | const double* inputVector, | ||
| 180 | unsigned int num_points) | ||
| 181 | { | ||
| 182 | 2 | float* outputVectorPtr = outputVector; | |
| 183 | 2 | const double* inputVectorPtr = inputVector; | |
| 184 | 2 | unsigned int number = 0; | |
| 185 | |||
| 186 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
| 187 | 262142 | *outputVectorPtr++ = ((float)(*inputVectorPtr++)); | |
| 188 | } | ||
| 189 | 2 | } | |
| 190 | #endif /* LV_HAVE_GENERIC */ | ||
| 191 | |||
| 192 | |||
| 193 | #endif /* INCLUDED_volk_64f_convert_32f_u_H */ | ||
| 194 | #ifndef INCLUDED_volk_64f_convert_32f_a_H | ||
| 195 | #define INCLUDED_volk_64f_convert_32f_a_H | ||
| 196 | |||
| 197 | #include <inttypes.h> | ||
| 198 | #include <stdio.h> | ||
| 199 | |||
| 200 | #ifdef LV_HAVE_AVX512F | ||
| 201 | #include <immintrin.h> | ||
| 202 | |||
| 203 | ✗ | static inline void volk_64f_convert_32f_a_avx512f(float* outputVector, | |
| 204 | const double* inputVector, | ||
| 205 | unsigned int num_points) | ||
| 206 | { | ||
| 207 | ✗ | unsigned int number = 0; | |
| 208 | |||
| 209 | ✗ | const unsigned int oneSixteenthPoints = num_points / 16; | |
| 210 | |||
| 211 | ✗ | const double* inputVectorPtr = (const double*)inputVector; | |
| 212 | ✗ | float* outputVectorPtr = outputVector; | |
| 213 | __m256 ret1, ret2; | ||
| 214 | __m512d inputVal1, inputVal2; | ||
| 215 | |||
| 216 | ✗ | for (; number < oneSixteenthPoints; number++) { | |
| 217 | ✗ | inputVal1 = _mm512_load_pd(inputVectorPtr); | |
| 218 | ✗ | inputVectorPtr += 8; | |
| 219 | ✗ | inputVal2 = _mm512_load_pd(inputVectorPtr); | |
| 220 | ✗ | inputVectorPtr += 8; | |
| 221 | |||
| 222 | ✗ | ret1 = _mm512_cvtpd_ps(inputVal1); | |
| 223 | ✗ | ret2 = _mm512_cvtpd_ps(inputVal2); | |
| 224 | |||
| 225 | _mm256_store_ps(outputVectorPtr, ret1); | ||
| 226 | ✗ | outputVectorPtr += 8; | |
| 227 | |||
| 228 | _mm256_store_ps(outputVectorPtr, ret2); | ||
| 229 | ✗ | outputVectorPtr += 8; | |
| 230 | } | ||
| 231 | |||
| 232 | ✗ | number = oneSixteenthPoints * 16; | |
| 233 | ✗ | for (; number < num_points; number++) { | |
| 234 | ✗ | outputVector[number] = (float)(inputVector[number]); | |
| 235 | } | ||
| 236 | ✗ | } | |
| 237 | #endif /* LV_HAVE_AVX512F */ | ||
| 238 | |||
| 239 | |||
| 240 | #ifdef LV_HAVE_AVX | ||
| 241 | #include <immintrin.h> | ||
| 242 | |||
| 243 | 2 | static inline void volk_64f_convert_32f_a_avx(float* outputVector, | |
| 244 | const double* inputVector, | ||
| 245 | unsigned int num_points) | ||
| 246 | { | ||
| 247 | 2 | unsigned int number = 0; | |
| 248 | |||
| 249 | 2 | const unsigned int oneEightPoints = num_points / 8; | |
| 250 | |||
| 251 | 2 | const double* inputVectorPtr = (const double*)inputVector; | |
| 252 | 2 | float* outputVectorPtr = outputVector; | |
| 253 | __m128 ret1, ret2; | ||
| 254 | __m256d inputVal1, inputVal2; | ||
| 255 | |||
| 256 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < oneEightPoints; number++) { |
| 257 | 32766 | inputVal1 = _mm256_load_pd(inputVectorPtr); | |
| 258 | 32766 | inputVectorPtr += 4; | |
| 259 | 32766 | inputVal2 = _mm256_load_pd(inputVectorPtr); | |
| 260 | 32766 | inputVectorPtr += 4; | |
| 261 | |||
| 262 | 32766 | ret1 = _mm256_cvtpd_ps(inputVal1); | |
| 263 | 32766 | ret2 = _mm256_cvtpd_ps(inputVal2); | |
| 264 | |||
| 265 | _mm_store_ps(outputVectorPtr, ret1); | ||
| 266 | 32766 | outputVectorPtr += 4; | |
| 267 | |||
| 268 | _mm_store_ps(outputVectorPtr, ret2); | ||
| 269 | 32766 | outputVectorPtr += 4; | |
| 270 | } | ||
| 271 | |||
| 272 | 2 | number = oneEightPoints * 8; | |
| 273 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 274 | 14 | outputVector[number] = (float)(inputVector[number]); | |
| 275 | } | ||
| 276 | 2 | } | |
| 277 | #endif /* LV_HAVE_AVX */ | ||
| 278 | |||
| 279 | |||
| 280 | #ifdef LV_HAVE_SSE2 | ||
| 281 | #include <emmintrin.h> | ||
| 282 | |||
| 283 | 2 | static inline void volk_64f_convert_32f_a_sse2(float* outputVector, | |
| 284 | const double* inputVector, | ||
| 285 | unsigned int num_points) | ||
| 286 | { | ||
| 287 | 2 | unsigned int number = 0; | |
| 288 | |||
| 289 | 2 | const unsigned int quarterPoints = num_points / 4; | |
| 290 | |||
| 291 | 2 | const double* inputVectorPtr = (const double*)inputVector; | |
| 292 | 2 | float* outputVectorPtr = outputVector; | |
| 293 | __m128 ret, ret2; | ||
| 294 | __m128d inputVal1, inputVal2; | ||
| 295 | |||
| 296 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 297 | 65534 | inputVal1 = _mm_load_pd(inputVectorPtr); | |
| 298 | 65534 | inputVectorPtr += 2; | |
| 299 | 65534 | inputVal2 = _mm_load_pd(inputVectorPtr); | |
| 300 | 65534 | inputVectorPtr += 2; | |
| 301 | |||
| 302 | 65534 | ret = _mm_cvtpd_ps(inputVal1); | |
| 303 | 65534 | ret2 = _mm_cvtpd_ps(inputVal2); | |
| 304 | |||
| 305 | 65534 | ret = _mm_movelh_ps(ret, ret2); | |
| 306 | |||
| 307 | _mm_store_ps(outputVectorPtr, ret); | ||
| 308 | 65534 | outputVectorPtr += 4; | |
| 309 | } | ||
| 310 | |||
| 311 | 2 | number = quarterPoints * 4; | |
| 312 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 313 | 6 | outputVector[number] = (float)(inputVector[number]); | |
| 314 | } | ||
| 315 | 2 | } | |
| 316 | #endif /* LV_HAVE_SSE2 */ | ||
| 317 | |||
| 318 | |||
| 319 | #ifdef LV_HAVE_GENERIC | ||
| 320 | |||
| 321 | 2 | static inline void volk_64f_convert_32f_a_generic(float* outputVector, | |
| 322 | const double* inputVector, | ||
| 323 | unsigned int num_points) | ||
| 324 | { | ||
| 325 | 2 | float* outputVectorPtr = outputVector; | |
| 326 | 2 | const double* inputVectorPtr = inputVector; | |
| 327 | 2 | unsigned int number = 0; | |
| 328 | |||
| 329 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
| 330 | 262142 | *outputVectorPtr++ = ((float)(*inputVectorPtr++)); | |
| 331 | } | ||
| 332 | 2 | } | |
| 333 | #endif /* LV_HAVE_GENERIC */ | ||
| 334 | |||
| 335 | |||
| 336 | #endif /* INCLUDED_volk_64f_convert_32f_a_H */ | ||
| 337 |