| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* -*- c++ -*- */ | ||
| 2 | /* | ||
| 3 | * Copyright 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_asin_32f | ||
| 12 | * | ||
| 13 | * \b Overview | ||
| 14 | * | ||
| 15 | * Computes arcsine of input vector and stores results in output vector. | ||
| 16 | * | ||
| 17 | * <b>Dispatcher Prototype</b> | ||
| 18 | * \code | ||
| 19 | * void volk_32f_asin_32f(float* bVector, const float* aVector, unsigned int num_points) | ||
| 20 | * \endcode | ||
| 21 | * | ||
| 22 | * \b Inputs | ||
| 23 | * \li aVector: The input vector of floats. | ||
| 24 | * \li num_points: The number of data points. | ||
| 25 | * | ||
| 26 | * \b Outputs | ||
| 27 | * \li bVector: The vector where results will be stored. | ||
| 28 | * | ||
| 29 | * \b Example | ||
| 30 | * \code | ||
| 31 | * Calculate common angles around the top half of the unit circle. | ||
| 32 | * int N = 10; | ||
| 33 | * unsigned int alignment = volk_get_alignment(); | ||
| 34 | * float* in = (float*)volk_malloc(sizeof(float)*N, alignment); | ||
| 35 | * float* out = (float*)volk_malloc(sizeof(float)*N, alignment); | ||
| 36 | * | ||
| 37 | * in[0] = 0; | ||
| 38 | * in[1] = 0.5; | ||
| 39 | * in[2] = std::sqrt(2.f)/2.f; | ||
| 40 | * in[3] = std::sqrt(3.f)/2.f; | ||
| 41 | * in[4] = in[5] = 1; | ||
| 42 | * for(unsigned int ii = 6; ii < N; ++ii){ | ||
| 43 | * in[ii] = - in[N-ii-1]; | ||
| 44 | * } | ||
| 45 | * | ||
| 46 | * volk_32f_asin_32f(out, in, N); | ||
| 47 | * | ||
| 48 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 49 | * printf("asin(%1.3f) = %1.3f\n", in[ii], out[ii]); | ||
| 50 | * } | ||
| 51 | * | ||
| 52 | * volk_free(in); | ||
| 53 | * volk_free(out); | ||
| 54 | * \endcode | ||
| 55 | */ | ||
| 56 | |||
| 57 | #include <inttypes.h> | ||
| 58 | #include <math.h> | ||
| 59 | #include <stdio.h> | ||
| 60 | |||
| 61 | /* This is the number of terms of Taylor series to evaluate, increase this for more | ||
| 62 | * accuracy*/ | ||
| 63 | #define ASIN_TERMS 2 | ||
| 64 | |||
| 65 | #ifndef INCLUDED_volk_32f_asin_32f_a_H | ||
| 66 | #define INCLUDED_volk_32f_asin_32f_a_H | ||
| 67 | |||
| 68 | #if LV_HAVE_AVX2 && LV_HAVE_FMA | ||
| 69 | #include <immintrin.h> | ||
| 70 | |||
| 71 | 2 | static inline void volk_32f_asin_32f_a_avx2_fma(float* bVector, | |
| 72 | const float* aVector, | ||
| 73 | unsigned int num_points) | ||
| 74 | { | ||
| 75 | 2 | float* bPtr = bVector; | |
| 76 | 2 | const float* aPtr = aVector; | |
| 77 | |||
| 78 | 2 | unsigned int number = 0; | |
| 79 | 2 | unsigned int eighthPoints = num_points / 8; | |
| 80 | int i, j; | ||
| 81 | |||
| 82 | __m256 aVal, pio2, x, y, z, arcsine; | ||
| 83 | __m256 fzeroes, fones, ftwos, ffours, condition; | ||
| 84 | |||
| 85 | 2 | pio2 = _mm256_set1_ps(3.14159265358979323846 / 2); | |
| 86 | 2 | fzeroes = _mm256_setzero_ps(); | |
| 87 | 2 | fones = _mm256_set1_ps(1.0); | |
| 88 | 2 | ftwos = _mm256_set1_ps(2.0); | |
| 89 | 2 | ffours = _mm256_set1_ps(4.0); | |
| 90 | |||
| 91 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 92 | 32766 | aVal = _mm256_load_ps(aPtr); | |
| 93 | 131064 | aVal = _mm256_div_ps(aVal, | |
| 94 | _mm256_sqrt_ps(_mm256_mul_ps(_mm256_add_ps(fones, aVal), | ||
| 95 | _mm256_sub_ps(fones, aVal)))); | ||
| 96 | 32766 | z = aVal; | |
| 97 | 32766 | condition = _mm256_cmp_ps(z, fzeroes, _CMP_LT_OS); | |
| 98 | 65532 | z = _mm256_sub_ps(z, _mm256_and_ps(_mm256_mul_ps(z, ftwos), condition)); | |
| 99 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_LT_OS); | |
| 100 | 98298 | x = _mm256_add_ps( | |
| 101 | z, _mm256_and_ps(_mm256_sub_ps(_mm256_div_ps(fones, z), z), condition)); | ||
| 102 | |||
| 103 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (i = 0; i < 2; i++) { |
| 104 | 196596 | x = _mm256_add_ps(x, _mm256_sqrt_ps(_mm256_fmadd_ps(x, x, fones))); | |
| 105 | } | ||
| 106 | 32766 | x = _mm256_div_ps(fones, x); | |
| 107 | 32766 | y = fzeroes; | |
| 108 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (j = ASIN_TERMS - 1; j >= 0; j--) { |
| 109 | 196596 | y = _mm256_fmadd_ps( | |
| 110 | 65532 | y, _mm256_mul_ps(x, x), _mm256_set1_ps(pow(-1, j) / (2 * j + 1))); | |
| 111 | } | ||
| 112 | |||
| 113 | 32766 | y = _mm256_mul_ps(y, _mm256_mul_ps(x, ffours)); | |
| 114 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_GT_OS); | |
| 115 | |||
| 116 | 65532 | y = _mm256_add_ps(y, _mm256_and_ps(_mm256_fnmadd_ps(y, ftwos, pio2), condition)); | |
| 117 | 32766 | arcsine = y; | |
| 118 | 32766 | condition = _mm256_cmp_ps(aVal, fzeroes, _CMP_LT_OS); | |
| 119 | 98298 | arcsine = _mm256_sub_ps(arcsine, | |
| 120 | _mm256_and_ps(_mm256_mul_ps(arcsine, ftwos), condition)); | ||
| 121 | |||
| 122 | _mm256_store_ps(bPtr, arcsine); | ||
| 123 | 32766 | aPtr += 8; | |
| 124 | 32766 | bPtr += 8; | |
| 125 | } | ||
| 126 | |||
| 127 | 2 | number = eighthPoints * 8; | |
| 128 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 129 | 14 | *bPtr++ = asin(*aPtr++); | |
| 130 | } | ||
| 131 | 2 | } | |
| 132 | |||
| 133 | #endif /* LV_HAVE_AVX2 && LV_HAVE_FMA for aligned */ | ||
| 134 | |||
| 135 | |||
| 136 | #ifdef LV_HAVE_AVX | ||
| 137 | #include <immintrin.h> | ||
| 138 | |||
| 139 | static inline void | ||
| 140 | 2 | volk_32f_asin_32f_a_avx(float* bVector, const float* aVector, unsigned int num_points) | |
| 141 | { | ||
| 142 | 2 | float* bPtr = bVector; | |
| 143 | 2 | const float* aPtr = aVector; | |
| 144 | |||
| 145 | 2 | unsigned int number = 0; | |
| 146 | 2 | unsigned int eighthPoints = num_points / 8; | |
| 147 | int i, j; | ||
| 148 | |||
| 149 | __m256 aVal, pio2, x, y, z, arcsine; | ||
| 150 | __m256 fzeroes, fones, ftwos, ffours, condition; | ||
| 151 | |||
| 152 | 2 | pio2 = _mm256_set1_ps(3.14159265358979323846 / 2); | |
| 153 | 2 | fzeroes = _mm256_setzero_ps(); | |
| 154 | 2 | fones = _mm256_set1_ps(1.0); | |
| 155 | 2 | ftwos = _mm256_set1_ps(2.0); | |
| 156 | 2 | ffours = _mm256_set1_ps(4.0); | |
| 157 | |||
| 158 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 159 | 32766 | aVal = _mm256_load_ps(aPtr); | |
| 160 | 131064 | aVal = _mm256_div_ps(aVal, | |
| 161 | _mm256_sqrt_ps(_mm256_mul_ps(_mm256_add_ps(fones, aVal), | ||
| 162 | _mm256_sub_ps(fones, aVal)))); | ||
| 163 | 32766 | z = aVal; | |
| 164 | 32766 | condition = _mm256_cmp_ps(z, fzeroes, _CMP_LT_OS); | |
| 165 | 65532 | z = _mm256_sub_ps(z, _mm256_and_ps(_mm256_mul_ps(z, ftwos), condition)); | |
| 166 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_LT_OS); | |
| 167 | 98298 | x = _mm256_add_ps( | |
| 168 | z, _mm256_and_ps(_mm256_sub_ps(_mm256_div_ps(fones, z), z), condition)); | ||
| 169 | |||
| 170 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (i = 0; i < 2; i++) { |
| 171 | 262128 | x = _mm256_add_ps(x, | |
| 172 | _mm256_sqrt_ps(_mm256_add_ps(fones, _mm256_mul_ps(x, x)))); | ||
| 173 | } | ||
| 174 | 32766 | x = _mm256_div_ps(fones, x); | |
| 175 | 32766 | y = fzeroes; | |
| 176 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (j = ASIN_TERMS - 1; j >= 0; j--) { |
| 177 | 262128 | y = _mm256_add_ps(_mm256_mul_ps(y, _mm256_mul_ps(x, x)), | |
| 178 | 65532 | _mm256_set1_ps(pow(-1, j) / (2 * j + 1))); | |
| 179 | } | ||
| 180 | |||
| 181 | 32766 | y = _mm256_mul_ps(y, _mm256_mul_ps(x, ffours)); | |
| 182 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_GT_OS); | |
| 183 | |||
| 184 | 98298 | y = _mm256_add_ps( | |
| 185 | y, _mm256_and_ps(_mm256_sub_ps(pio2, _mm256_mul_ps(y, ftwos)), condition)); | ||
| 186 | 32766 | arcsine = y; | |
| 187 | 32766 | condition = _mm256_cmp_ps(aVal, fzeroes, _CMP_LT_OS); | |
| 188 | 98298 | arcsine = _mm256_sub_ps(arcsine, | |
| 189 | _mm256_and_ps(_mm256_mul_ps(arcsine, ftwos), condition)); | ||
| 190 | |||
| 191 | _mm256_store_ps(bPtr, arcsine); | ||
| 192 | 32766 | aPtr += 8; | |
| 193 | 32766 | bPtr += 8; | |
| 194 | } | ||
| 195 | |||
| 196 | 2 | number = eighthPoints * 8; | |
| 197 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 198 | 14 | *bPtr++ = asin(*aPtr++); | |
| 199 | } | ||
| 200 | 2 | } | |
| 201 | |||
| 202 | #endif /* LV_HAVE_AVX for aligned */ | ||
| 203 | |||
| 204 | #ifdef LV_HAVE_SSE4_1 | ||
| 205 | #include <smmintrin.h> | ||
| 206 | |||
| 207 | static inline void | ||
| 208 | 2 | volk_32f_asin_32f_a_sse4_1(float* bVector, const float* aVector, unsigned int num_points) | |
| 209 | { | ||
| 210 | 2 | float* bPtr = bVector; | |
| 211 | 2 | const float* aPtr = aVector; | |
| 212 | |||
| 213 | 2 | unsigned int number = 0; | |
| 214 | 2 | unsigned int quarterPoints = num_points / 4; | |
| 215 | int i, j; | ||
| 216 | |||
| 217 | __m128 aVal, pio2, x, y, z, arcsine; | ||
| 218 | __m128 fzeroes, fones, ftwos, ffours, condition; | ||
| 219 | |||
| 220 | 2 | pio2 = _mm_set1_ps(3.14159265358979323846 / 2); | |
| 221 | 2 | fzeroes = _mm_setzero_ps(); | |
| 222 | 2 | fones = _mm_set1_ps(1.0); | |
| 223 | 2 | ftwos = _mm_set1_ps(2.0); | |
| 224 | 2 | ffours = _mm_set1_ps(4.0); | |
| 225 | |||
| 226 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 227 | 65534 | aVal = _mm_load_ps(aPtr); | |
| 228 | 262136 | aVal = _mm_div_ps( | |
| 229 | aVal, | ||
| 230 | _mm_sqrt_ps(_mm_mul_ps(_mm_add_ps(fones, aVal), _mm_sub_ps(fones, aVal)))); | ||
| 231 | 65534 | z = aVal; | |
| 232 | 65534 | condition = _mm_cmplt_ps(z, fzeroes); | |
| 233 | 196602 | z = _mm_sub_ps(z, _mm_and_ps(_mm_mul_ps(z, ftwos), condition)); | |
| 234 | 65534 | condition = _mm_cmplt_ps(z, fones); | |
| 235 | 196602 | x = _mm_add_ps(z, _mm_and_ps(_mm_sub_ps(_mm_div_ps(fones, z), z), condition)); | |
| 236 | |||
| 237 |
2/2✓ Branch 0 taken 131068 times.
✓ Branch 1 taken 65534 times.
|
196602 | for (i = 0; i < 2; i++) { |
| 238 | 524272 | x = _mm_add_ps(x, _mm_sqrt_ps(_mm_add_ps(fones, _mm_mul_ps(x, x)))); | |
| 239 | } | ||
| 240 | 65534 | x = _mm_div_ps(fones, x); | |
| 241 | 65534 | y = fzeroes; | |
| 242 |
2/2✓ Branch 0 taken 131068 times.
✓ Branch 1 taken 65534 times.
|
196602 | for (j = ASIN_TERMS - 1; j >= 0; j--) { |
| 243 | 524272 | y = _mm_add_ps(_mm_mul_ps(y, _mm_mul_ps(x, x)), | |
| 244 | 131068 | _mm_set1_ps(pow(-1, j) / (2 * j + 1))); | |
| 245 | } | ||
| 246 | |||
| 247 | 131068 | y = _mm_mul_ps(y, _mm_mul_ps(x, ffours)); | |
| 248 | 65534 | condition = _mm_cmpgt_ps(z, fones); | |
| 249 | |||
| 250 | 196602 | y = _mm_add_ps(y, _mm_and_ps(_mm_sub_ps(pio2, _mm_mul_ps(y, ftwos)), condition)); | |
| 251 | 65534 | arcsine = y; | |
| 252 | 65534 | condition = _mm_cmplt_ps(aVal, fzeroes); | |
| 253 | 196602 | arcsine = _mm_sub_ps(arcsine, _mm_and_ps(_mm_mul_ps(arcsine, ftwos), condition)); | |
| 254 | |||
| 255 | _mm_store_ps(bPtr, arcsine); | ||
| 256 | 65534 | aPtr += 4; | |
| 257 | 65534 | bPtr += 4; | |
| 258 | } | ||
| 259 | |||
| 260 | 2 | number = quarterPoints * 4; | |
| 261 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 262 | 6 | *bPtr++ = asinf(*aPtr++); | |
| 263 | } | ||
| 264 | 2 | } | |
| 265 | |||
| 266 | #endif /* LV_HAVE_SSE4_1 for aligned */ | ||
| 267 | |||
| 268 | #endif /* INCLUDED_volk_32f_asin_32f_a_H */ | ||
| 269 | |||
| 270 | #ifndef INCLUDED_volk_32f_asin_32f_u_H | ||
| 271 | #define INCLUDED_volk_32f_asin_32f_u_H | ||
| 272 | |||
| 273 | #if LV_HAVE_AVX2 && LV_HAVE_FMA | ||
| 274 | #include <immintrin.h> | ||
| 275 | |||
| 276 | 2 | static inline void volk_32f_asin_32f_u_avx2_fma(float* bVector, | |
| 277 | const float* aVector, | ||
| 278 | unsigned int num_points) | ||
| 279 | { | ||
| 280 | 2 | float* bPtr = bVector; | |
| 281 | 2 | const float* aPtr = aVector; | |
| 282 | |||
| 283 | 2 | unsigned int number = 0; | |
| 284 | 2 | unsigned int eighthPoints = num_points / 8; | |
| 285 | int i, j; | ||
| 286 | |||
| 287 | __m256 aVal, pio2, x, y, z, arcsine; | ||
| 288 | __m256 fzeroes, fones, ftwos, ffours, condition; | ||
| 289 | |||
| 290 | 2 | pio2 = _mm256_set1_ps(3.14159265358979323846 / 2); | |
| 291 | 2 | fzeroes = _mm256_setzero_ps(); | |
| 292 | 2 | fones = _mm256_set1_ps(1.0); | |
| 293 | 2 | ftwos = _mm256_set1_ps(2.0); | |
| 294 | 2 | ffours = _mm256_set1_ps(4.0); | |
| 295 | |||
| 296 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 297 | 32766 | aVal = _mm256_loadu_ps(aPtr); | |
| 298 | 131064 | aVal = _mm256_div_ps(aVal, | |
| 299 | _mm256_sqrt_ps(_mm256_mul_ps(_mm256_add_ps(fones, aVal), | ||
| 300 | _mm256_sub_ps(fones, aVal)))); | ||
| 301 | 32766 | z = aVal; | |
| 302 | 32766 | condition = _mm256_cmp_ps(z, fzeroes, _CMP_LT_OS); | |
| 303 | 65532 | z = _mm256_sub_ps(z, _mm256_and_ps(_mm256_mul_ps(z, ftwos), condition)); | |
| 304 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_LT_OS); | |
| 305 | 98298 | x = _mm256_add_ps( | |
| 306 | z, _mm256_and_ps(_mm256_sub_ps(_mm256_div_ps(fones, z), z), condition)); | ||
| 307 | |||
| 308 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (i = 0; i < 2; i++) { |
| 309 | 196596 | x = _mm256_add_ps(x, _mm256_sqrt_ps(_mm256_fmadd_ps(x, x, fones))); | |
| 310 | } | ||
| 311 | 32766 | x = _mm256_div_ps(fones, x); | |
| 312 | 32766 | y = fzeroes; | |
| 313 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (j = ASIN_TERMS - 1; j >= 0; j--) { |
| 314 | 196596 | y = _mm256_fmadd_ps( | |
| 315 | 65532 | y, _mm256_mul_ps(x, x), _mm256_set1_ps(pow(-1, j) / (2 * j + 1))); | |
| 316 | } | ||
| 317 | |||
| 318 | 32766 | y = _mm256_mul_ps(y, _mm256_mul_ps(x, ffours)); | |
| 319 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_GT_OS); | |
| 320 | |||
| 321 | 65532 | y = _mm256_add_ps(y, _mm256_and_ps(_mm256_fnmadd_ps(y, ftwos, pio2), condition)); | |
| 322 | 32766 | arcsine = y; | |
| 323 | 32766 | condition = _mm256_cmp_ps(aVal, fzeroes, _CMP_LT_OS); | |
| 324 | 98298 | arcsine = _mm256_sub_ps(arcsine, | |
| 325 | _mm256_and_ps(_mm256_mul_ps(arcsine, ftwos), condition)); | ||
| 326 | |||
| 327 | _mm256_storeu_ps(bPtr, arcsine); | ||
| 328 | 32766 | aPtr += 8; | |
| 329 | 32766 | bPtr += 8; | |
| 330 | } | ||
| 331 | |||
| 332 | 2 | number = eighthPoints * 8; | |
| 333 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 334 | 14 | *bPtr++ = asin(*aPtr++); | |
| 335 | } | ||
| 336 | 2 | } | |
| 337 | |||
| 338 | #endif /* LV_HAVE_AVX2 && LV_HAVE_FMA for unaligned */ | ||
| 339 | |||
| 340 | |||
| 341 | #ifdef LV_HAVE_AVX | ||
| 342 | #include <immintrin.h> | ||
| 343 | |||
| 344 | static inline void | ||
| 345 | 2 | volk_32f_asin_32f_u_avx(float* bVector, const float* aVector, unsigned int num_points) | |
| 346 | { | ||
| 347 | 2 | float* bPtr = bVector; | |
| 348 | 2 | const float* aPtr = aVector; | |
| 349 | |||
| 350 | 2 | unsigned int number = 0; | |
| 351 | 2 | unsigned int eighthPoints = num_points / 8; | |
| 352 | int i, j; | ||
| 353 | |||
| 354 | __m256 aVal, pio2, x, y, z, arcsine; | ||
| 355 | __m256 fzeroes, fones, ftwos, ffours, condition; | ||
| 356 | |||
| 357 | 2 | pio2 = _mm256_set1_ps(3.14159265358979323846 / 2); | |
| 358 | 2 | fzeroes = _mm256_setzero_ps(); | |
| 359 | 2 | fones = _mm256_set1_ps(1.0); | |
| 360 | 2 | ftwos = _mm256_set1_ps(2.0); | |
| 361 | 2 | ffours = _mm256_set1_ps(4.0); | |
| 362 | |||
| 363 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 364 | 32766 | aVal = _mm256_loadu_ps(aPtr); | |
| 365 | 131064 | aVal = _mm256_div_ps(aVal, | |
| 366 | _mm256_sqrt_ps(_mm256_mul_ps(_mm256_add_ps(fones, aVal), | ||
| 367 | _mm256_sub_ps(fones, aVal)))); | ||
| 368 | 32766 | z = aVal; | |
| 369 | 32766 | condition = _mm256_cmp_ps(z, fzeroes, _CMP_LT_OS); | |
| 370 | 65532 | z = _mm256_sub_ps(z, _mm256_and_ps(_mm256_mul_ps(z, ftwos), condition)); | |
| 371 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_LT_OS); | |
| 372 | 98298 | x = _mm256_add_ps( | |
| 373 | z, _mm256_and_ps(_mm256_sub_ps(_mm256_div_ps(fones, z), z), condition)); | ||
| 374 | |||
| 375 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (i = 0; i < 2; i++) { |
| 376 | 262128 | x = _mm256_add_ps(x, | |
| 377 | _mm256_sqrt_ps(_mm256_add_ps(fones, _mm256_mul_ps(x, x)))); | ||
| 378 | } | ||
| 379 | 32766 | x = _mm256_div_ps(fones, x); | |
| 380 | 32766 | y = fzeroes; | |
| 381 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (j = ASIN_TERMS - 1; j >= 0; j--) { |
| 382 | 262128 | y = _mm256_add_ps(_mm256_mul_ps(y, _mm256_mul_ps(x, x)), | |
| 383 | 65532 | _mm256_set1_ps(pow(-1, j) / (2 * j + 1))); | |
| 384 | } | ||
| 385 | |||
| 386 | 32766 | y = _mm256_mul_ps(y, _mm256_mul_ps(x, ffours)); | |
| 387 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_GT_OS); | |
| 388 | |||
| 389 | 98298 | y = _mm256_add_ps( | |
| 390 | y, _mm256_and_ps(_mm256_sub_ps(pio2, _mm256_mul_ps(y, ftwos)), condition)); | ||
| 391 | 32766 | arcsine = y; | |
| 392 | 32766 | condition = _mm256_cmp_ps(aVal, fzeroes, _CMP_LT_OS); | |
| 393 | 98298 | arcsine = _mm256_sub_ps(arcsine, | |
| 394 | _mm256_and_ps(_mm256_mul_ps(arcsine, ftwos), condition)); | ||
| 395 | |||
| 396 | _mm256_storeu_ps(bPtr, arcsine); | ||
| 397 | 32766 | aPtr += 8; | |
| 398 | 32766 | bPtr += 8; | |
| 399 | } | ||
| 400 | |||
| 401 | 2 | number = eighthPoints * 8; | |
| 402 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 403 | 14 | *bPtr++ = asin(*aPtr++); | |
| 404 | } | ||
| 405 | 2 | } | |
| 406 | |||
| 407 | #endif /* LV_HAVE_AVX for unaligned */ | ||
| 408 | |||
| 409 | |||
| 410 | #ifdef LV_HAVE_SSE4_1 | ||
| 411 | #include <smmintrin.h> | ||
| 412 | |||
| 413 | static inline void | ||
| 414 | 2 | volk_32f_asin_32f_u_sse4_1(float* bVector, const float* aVector, unsigned int num_points) | |
| 415 | { | ||
| 416 | 2 | float* bPtr = bVector; | |
| 417 | 2 | const float* aPtr = aVector; | |
| 418 | |||
| 419 | 2 | unsigned int number = 0; | |
| 420 | 2 | unsigned int quarterPoints = num_points / 4; | |
| 421 | int i, j; | ||
| 422 | |||
| 423 | __m128 aVal, pio2, x, y, z, arcsine; | ||
| 424 | __m128 fzeroes, fones, ftwos, ffours, condition; | ||
| 425 | |||
| 426 | 2 | pio2 = _mm_set1_ps(3.14159265358979323846 / 2); | |
| 427 | 2 | fzeroes = _mm_setzero_ps(); | |
| 428 | 2 | fones = _mm_set1_ps(1.0); | |
| 429 | 2 | ftwos = _mm_set1_ps(2.0); | |
| 430 | 2 | ffours = _mm_set1_ps(4.0); | |
| 431 | |||
| 432 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 433 | 65534 | aVal = _mm_loadu_ps(aPtr); | |
| 434 | 262136 | aVal = _mm_div_ps( | |
| 435 | aVal, | ||
| 436 | _mm_sqrt_ps(_mm_mul_ps(_mm_add_ps(fones, aVal), _mm_sub_ps(fones, aVal)))); | ||
| 437 | 65534 | z = aVal; | |
| 438 | 65534 | condition = _mm_cmplt_ps(z, fzeroes); | |
| 439 | 196602 | z = _mm_sub_ps(z, _mm_and_ps(_mm_mul_ps(z, ftwos), condition)); | |
| 440 | 65534 | condition = _mm_cmplt_ps(z, fones); | |
| 441 | 196602 | x = _mm_add_ps(z, _mm_and_ps(_mm_sub_ps(_mm_div_ps(fones, z), z), condition)); | |
| 442 | |||
| 443 |
2/2✓ Branch 0 taken 131068 times.
✓ Branch 1 taken 65534 times.
|
196602 | for (i = 0; i < 2; i++) { |
| 444 | 524272 | x = _mm_add_ps(x, _mm_sqrt_ps(_mm_add_ps(fones, _mm_mul_ps(x, x)))); | |
| 445 | } | ||
| 446 | 65534 | x = _mm_div_ps(fones, x); | |
| 447 | 65534 | y = fzeroes; | |
| 448 |
2/2✓ Branch 0 taken 131068 times.
✓ Branch 1 taken 65534 times.
|
196602 | for (j = ASIN_TERMS - 1; j >= 0; j--) { |
| 449 | 524272 | y = _mm_add_ps(_mm_mul_ps(y, _mm_mul_ps(x, x)), | |
| 450 | 131068 | _mm_set1_ps(pow(-1, j) / (2 * j + 1))); | |
| 451 | } | ||
| 452 | |||
| 453 | 131068 | y = _mm_mul_ps(y, _mm_mul_ps(x, ffours)); | |
| 454 | 65534 | condition = _mm_cmpgt_ps(z, fones); | |
| 455 | |||
| 456 | 196602 | y = _mm_add_ps(y, _mm_and_ps(_mm_sub_ps(pio2, _mm_mul_ps(y, ftwos)), condition)); | |
| 457 | 65534 | arcsine = y; | |
| 458 | 65534 | condition = _mm_cmplt_ps(aVal, fzeroes); | |
| 459 | 196602 | arcsine = _mm_sub_ps(arcsine, _mm_and_ps(_mm_mul_ps(arcsine, ftwos), condition)); | |
| 460 | |||
| 461 | _mm_storeu_ps(bPtr, arcsine); | ||
| 462 | 65534 | aPtr += 4; | |
| 463 | 65534 | bPtr += 4; | |
| 464 | } | ||
| 465 | |||
| 466 | 2 | number = quarterPoints * 4; | |
| 467 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 468 | 6 | *bPtr++ = asinf(*aPtr++); | |
| 469 | } | ||
| 470 | 2 | } | |
| 471 | |||
| 472 | #endif /* LV_HAVE_SSE4_1 for unaligned */ | ||
| 473 | |||
| 474 | #ifdef LV_HAVE_GENERIC | ||
| 475 | |||
| 476 | static inline void | ||
| 477 | 2 | volk_32f_asin_32f_u_generic(float* bVector, const float* aVector, unsigned int num_points) | |
| 478 | { | ||
| 479 | 2 | float* bPtr = bVector; | |
| 480 | 2 | const float* aPtr = aVector; | |
| 481 | 2 | unsigned int number = 0; | |
| 482 | |||
| 483 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
| 484 | 262142 | *bPtr++ = asinf(*aPtr++); | |
| 485 | } | ||
| 486 | 2 | } | |
| 487 | #endif /* LV_HAVE_GENERIC */ | ||
| 488 | |||
| 489 | #endif /* INCLUDED_volk_32f_asin_32f_u_H */ | ||
| 490 |