| 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_acos_32f | ||
| 12 | * | ||
| 13 | * \b Overview | ||
| 14 | * | ||
| 15 | * Computes arccosine of the input vector and stores results in the output vector. | ||
| 16 | * | ||
| 17 | * <b>Dispatcher Prototype</b> | ||
| 18 | * \code | ||
| 19 | * void volk_32f_acos_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 | * Calculate common angles around the top half of the unit circle. | ||
| 31 | * \code | ||
| 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] = 1; | ||
| 38 | * in[1] = std::sqrt(3.f)/2.f; | ||
| 39 | * in[2] = std::sqrt(2.f)/2.f; | ||
| 40 | * in[3] = 0.5; | ||
| 41 | * in[4] = in[5] = 0; | ||
| 42 | * for(unsigned int ii = 6; ii < N; ++ii){ | ||
| 43 | * in[ii] = - in[N-ii-1]; | ||
| 44 | * } | ||
| 45 | * | ||
| 46 | * volk_32f_acos_32f(out, in, N); | ||
| 47 | * | ||
| 48 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 49 | * printf("acos(%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 ACOS_TERMS 2 | ||
| 64 | |||
| 65 | #ifndef INCLUDED_volk_32f_acos_32f_a_H | ||
| 66 | #define INCLUDED_volk_32f_acos_32f_a_H | ||
| 67 | |||
| 68 | #if LV_HAVE_AVX2 && LV_HAVE_FMA | ||
| 69 | #include <immintrin.h> | ||
| 70 | |||
| 71 | 2 | static inline void volk_32f_acos_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, d, pi, pio2, x, y, z, arccosine; | ||
| 83 | __m256 fzeroes, fones, ftwos, ffours, condition; | ||
| 84 | |||
| 85 | 2 | pi = _mm256_set1_ps(3.14159265358979323846); | |
| 86 | 2 | pio2 = _mm256_set1_ps(3.14159265358979323846 / 2); | |
| 87 | 2 | fzeroes = _mm256_setzero_ps(); | |
| 88 | 2 | fones = _mm256_set1_ps(1.0); | |
| 89 | 2 | ftwos = _mm256_set1_ps(2.0); | |
| 90 | 2 | ffours = _mm256_set1_ps(4.0); | |
| 91 | |||
| 92 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 93 | 32766 | aVal = _mm256_load_ps(aPtr); | |
| 94 | 32766 | d = aVal; | |
| 95 | 131064 | aVal = _mm256_div_ps(_mm256_sqrt_ps(_mm256_mul_ps(_mm256_add_ps(fones, aVal), | |
| 96 | _mm256_sub_ps(fones, aVal))), | ||
| 97 | aVal); | ||
| 98 | 32766 | z = aVal; | |
| 99 | 32766 | condition = _mm256_cmp_ps(z, fzeroes, _CMP_LT_OS); | |
| 100 | 65532 | z = _mm256_sub_ps(z, _mm256_and_ps(_mm256_mul_ps(z, ftwos), condition)); | |
| 101 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_LT_OS); | |
| 102 | 98298 | x = _mm256_add_ps( | |
| 103 | z, _mm256_and_ps(_mm256_sub_ps(_mm256_div_ps(fones, z), z), condition)); | ||
| 104 | |||
| 105 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (i = 0; i < 2; i++) |
| 106 | 196596 | x = _mm256_add_ps(x, _mm256_sqrt_ps(_mm256_fmadd_ps(x, x, fones))); | |
| 107 | 32766 | x = _mm256_div_ps(fones, x); | |
| 108 | 32766 | y = fzeroes; | |
| 109 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (j = ACOS_TERMS - 1; j >= 0; j--) |
| 110 | 196596 | y = _mm256_fmadd_ps( | |
| 111 | 65532 | y, _mm256_mul_ps(x, x), _mm256_set1_ps(pow(-1, j) / (2 * j + 1))); | |
| 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 | arccosine = y; | |
| 118 | 32766 | condition = _mm256_cmp_ps(aVal, fzeroes, _CMP_LT_OS); | |
| 119 | 65532 | arccosine = _mm256_sub_ps( | |
| 120 | arccosine, _mm256_and_ps(_mm256_mul_ps(arccosine, ftwos), condition)); | ||
| 121 | 32766 | condition = _mm256_cmp_ps(d, fzeroes, _CMP_LT_OS); | |
| 122 | 65532 | arccosine = _mm256_add_ps(arccosine, _mm256_and_ps(pi, condition)); | |
| 123 | |||
| 124 | _mm256_store_ps(bPtr, arccosine); | ||
| 125 | 32766 | aPtr += 8; | |
| 126 | 32766 | bPtr += 8; | |
| 127 | } | ||
| 128 | |||
| 129 | 2 | number = eighthPoints * 8; | |
| 130 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 131 | 14 | *bPtr++ = acos(*aPtr++); | |
| 132 | } | ||
| 133 | 2 | } | |
| 134 | |||
| 135 | #endif /* LV_HAVE_AVX2 && LV_HAVE_FMA for aligned */ | ||
| 136 | |||
| 137 | |||
| 138 | #ifdef LV_HAVE_AVX | ||
| 139 | #include <immintrin.h> | ||
| 140 | |||
| 141 | static inline void | ||
| 142 | 2 | volk_32f_acos_32f_a_avx(float* bVector, const float* aVector, unsigned int num_points) | |
| 143 | { | ||
| 144 | 2 | float* bPtr = bVector; | |
| 145 | 2 | const float* aPtr = aVector; | |
| 146 | |||
| 147 | 2 | unsigned int number = 0; | |
| 148 | 2 | unsigned int eighthPoints = num_points / 8; | |
| 149 | int i, j; | ||
| 150 | |||
| 151 | __m256 aVal, d, pi, pio2, x, y, z, arccosine; | ||
| 152 | __m256 fzeroes, fones, ftwos, ffours, condition; | ||
| 153 | |||
| 154 | 2 | pi = _mm256_set1_ps(3.14159265358979323846); | |
| 155 | 2 | pio2 = _mm256_set1_ps(3.14159265358979323846 / 2); | |
| 156 | 2 | fzeroes = _mm256_setzero_ps(); | |
| 157 | 2 | fones = _mm256_set1_ps(1.0); | |
| 158 | 2 | ftwos = _mm256_set1_ps(2.0); | |
| 159 | 2 | ffours = _mm256_set1_ps(4.0); | |
| 160 | |||
| 161 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 162 | 32766 | aVal = _mm256_load_ps(aPtr); | |
| 163 | 32766 | d = aVal; | |
| 164 | 131064 | aVal = _mm256_div_ps(_mm256_sqrt_ps(_mm256_mul_ps(_mm256_add_ps(fones, aVal), | |
| 165 | _mm256_sub_ps(fones, aVal))), | ||
| 166 | aVal); | ||
| 167 | 32766 | z = aVal; | |
| 168 | 32766 | condition = _mm256_cmp_ps(z, fzeroes, _CMP_LT_OS); | |
| 169 | 65532 | z = _mm256_sub_ps(z, _mm256_and_ps(_mm256_mul_ps(z, ftwos), condition)); | |
| 170 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_LT_OS); | |
| 171 | 98298 | x = _mm256_add_ps( | |
| 172 | z, _mm256_and_ps(_mm256_sub_ps(_mm256_div_ps(fones, z), z), condition)); | ||
| 173 | |||
| 174 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (i = 0; i < 2; i++) |
| 175 | 262128 | x = _mm256_add_ps(x, | |
| 176 | _mm256_sqrt_ps(_mm256_add_ps(fones, _mm256_mul_ps(x, x)))); | ||
| 177 | 32766 | x = _mm256_div_ps(fones, x); | |
| 178 | 32766 | y = fzeroes; | |
| 179 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (j = ACOS_TERMS - 1; j >= 0; j--) |
| 180 | 262128 | y = _mm256_add_ps(_mm256_mul_ps(y, _mm256_mul_ps(x, x)), | |
| 181 | 65532 | _mm256_set1_ps(pow(-1, j) / (2 * j + 1))); | |
| 182 | |||
| 183 | 32766 | y = _mm256_mul_ps(y, _mm256_mul_ps(x, ffours)); | |
| 184 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_GT_OS); | |
| 185 | |||
| 186 | 98298 | y = _mm256_add_ps( | |
| 187 | y, _mm256_and_ps(_mm256_sub_ps(pio2, _mm256_mul_ps(y, ftwos)), condition)); | ||
| 188 | 32766 | arccosine = y; | |
| 189 | 32766 | condition = _mm256_cmp_ps(aVal, fzeroes, _CMP_LT_OS); | |
| 190 | 65532 | arccosine = _mm256_sub_ps( | |
| 191 | arccosine, _mm256_and_ps(_mm256_mul_ps(arccosine, ftwos), condition)); | ||
| 192 | 32766 | condition = _mm256_cmp_ps(d, fzeroes, _CMP_LT_OS); | |
| 193 | 65532 | arccosine = _mm256_add_ps(arccosine, _mm256_and_ps(pi, condition)); | |
| 194 | |||
| 195 | _mm256_store_ps(bPtr, arccosine); | ||
| 196 | 32766 | aPtr += 8; | |
| 197 | 32766 | bPtr += 8; | |
| 198 | } | ||
| 199 | |||
| 200 | 2 | number = eighthPoints * 8; | |
| 201 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 202 | 14 | *bPtr++ = acos(*aPtr++); | |
| 203 | } | ||
| 204 | 2 | } | |
| 205 | |||
| 206 | #endif /* LV_HAVE_AVX2 for aligned */ | ||
| 207 | |||
| 208 | #ifdef LV_HAVE_SSE4_1 | ||
| 209 | #include <smmintrin.h> | ||
| 210 | |||
| 211 | static inline void | ||
| 212 | 2 | volk_32f_acos_32f_a_sse4_1(float* bVector, const float* aVector, unsigned int num_points) | |
| 213 | { | ||
| 214 | 2 | float* bPtr = bVector; | |
| 215 | 2 | const float* aPtr = aVector; | |
| 216 | |||
| 217 | 2 | unsigned int number = 0; | |
| 218 | 2 | unsigned int quarterPoints = num_points / 4; | |
| 219 | int i, j; | ||
| 220 | |||
| 221 | __m128 aVal, d, pi, pio2, x, y, z, arccosine; | ||
| 222 | __m128 fzeroes, fones, ftwos, ffours, condition; | ||
| 223 | |||
| 224 | 2 | pi = _mm_set1_ps(3.14159265358979323846); | |
| 225 | 2 | pio2 = _mm_set1_ps(3.14159265358979323846 / 2); | |
| 226 | 2 | fzeroes = _mm_setzero_ps(); | |
| 227 | 2 | fones = _mm_set1_ps(1.0); | |
| 228 | 2 | ftwos = _mm_set1_ps(2.0); | |
| 229 | 2 | ffours = _mm_set1_ps(4.0); | |
| 230 | |||
| 231 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 232 | 65534 | aVal = _mm_load_ps(aPtr); | |
| 233 | 65534 | d = aVal; | |
| 234 | 262136 | aVal = _mm_div_ps( | |
| 235 | _mm_sqrt_ps(_mm_mul_ps(_mm_add_ps(fones, aVal), _mm_sub_ps(fones, aVal))), | ||
| 236 | aVal); | ||
| 237 | 65534 | z = aVal; | |
| 238 | 65534 | condition = _mm_cmplt_ps(z, fzeroes); | |
| 239 | 196602 | z = _mm_sub_ps(z, _mm_and_ps(_mm_mul_ps(z, ftwos), condition)); | |
| 240 | 65534 | condition = _mm_cmplt_ps(z, fones); | |
| 241 | 196602 | x = _mm_add_ps(z, _mm_and_ps(_mm_sub_ps(_mm_div_ps(fones, z), z), condition)); | |
| 242 | |||
| 243 |
2/2✓ Branch 0 taken 131068 times.
✓ Branch 1 taken 65534 times.
|
196602 | for (i = 0; i < 2; i++) |
| 244 | 524272 | x = _mm_add_ps(x, _mm_sqrt_ps(_mm_add_ps(fones, _mm_mul_ps(x, x)))); | |
| 245 | 65534 | x = _mm_div_ps(fones, x); | |
| 246 | 65534 | y = fzeroes; | |
| 247 |
2/2✓ Branch 0 taken 131068 times.
✓ Branch 1 taken 65534 times.
|
196602 | for (j = ACOS_TERMS - 1; j >= 0; j--) |
| 248 | 524272 | y = _mm_add_ps(_mm_mul_ps(y, _mm_mul_ps(x, x)), | |
| 249 | 131068 | _mm_set1_ps(pow(-1, j) / (2 * j + 1))); | |
| 250 | |||
| 251 | 131068 | y = _mm_mul_ps(y, _mm_mul_ps(x, ffours)); | |
| 252 | 65534 | condition = _mm_cmpgt_ps(z, fones); | |
| 253 | |||
| 254 | 196602 | y = _mm_add_ps(y, _mm_and_ps(_mm_sub_ps(pio2, _mm_mul_ps(y, ftwos)), condition)); | |
| 255 | 65534 | arccosine = y; | |
| 256 | 65534 | condition = _mm_cmplt_ps(aVal, fzeroes); | |
| 257 | arccosine = | ||
| 258 | 196602 | _mm_sub_ps(arccosine, _mm_and_ps(_mm_mul_ps(arccosine, ftwos), condition)); | |
| 259 | 65534 | condition = _mm_cmplt_ps(d, fzeroes); | |
| 260 | 131068 | arccosine = _mm_add_ps(arccosine, _mm_and_ps(pi, condition)); | |
| 261 | |||
| 262 | _mm_store_ps(bPtr, arccosine); | ||
| 263 | 65534 | aPtr += 4; | |
| 264 | 65534 | bPtr += 4; | |
| 265 | } | ||
| 266 | |||
| 267 | 2 | number = quarterPoints * 4; | |
| 268 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 269 | 6 | *bPtr++ = acosf(*aPtr++); | |
| 270 | } | ||
| 271 | 2 | } | |
| 272 | |||
| 273 | #endif /* LV_HAVE_SSE4_1 for aligned */ | ||
| 274 | |||
| 275 | #endif /* INCLUDED_volk_32f_acos_32f_a_H */ | ||
| 276 | |||
| 277 | |||
| 278 | #ifndef INCLUDED_volk_32f_acos_32f_u_H | ||
| 279 | #define INCLUDED_volk_32f_acos_32f_u_H | ||
| 280 | |||
| 281 | #if LV_HAVE_AVX2 && LV_HAVE_FMA | ||
| 282 | #include <immintrin.h> | ||
| 283 | |||
| 284 | 2 | static inline void volk_32f_acos_32f_u_avx2_fma(float* bVector, | |
| 285 | const float* aVector, | ||
| 286 | unsigned int num_points) | ||
| 287 | { | ||
| 288 | 2 | float* bPtr = bVector; | |
| 289 | 2 | const float* aPtr = aVector; | |
| 290 | |||
| 291 | 2 | unsigned int number = 0; | |
| 292 | 2 | unsigned int eighthPoints = num_points / 8; | |
| 293 | int i, j; | ||
| 294 | |||
| 295 | __m256 aVal, d, pi, pio2, x, y, z, arccosine; | ||
| 296 | __m256 fzeroes, fones, ftwos, ffours, condition; | ||
| 297 | |||
| 298 | 2 | pi = _mm256_set1_ps(3.14159265358979323846); | |
| 299 | 2 | pio2 = _mm256_set1_ps(3.14159265358979323846 / 2); | |
| 300 | 2 | fzeroes = _mm256_setzero_ps(); | |
| 301 | 2 | fones = _mm256_set1_ps(1.0); | |
| 302 | 2 | ftwos = _mm256_set1_ps(2.0); | |
| 303 | 2 | ffours = _mm256_set1_ps(4.0); | |
| 304 | |||
| 305 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 306 | 32766 | aVal = _mm256_loadu_ps(aPtr); | |
| 307 | 32766 | d = aVal; | |
| 308 | 131064 | aVal = _mm256_div_ps(_mm256_sqrt_ps(_mm256_mul_ps(_mm256_add_ps(fones, aVal), | |
| 309 | _mm256_sub_ps(fones, aVal))), | ||
| 310 | aVal); | ||
| 311 | 32766 | z = aVal; | |
| 312 | 32766 | condition = _mm256_cmp_ps(z, fzeroes, _CMP_LT_OS); | |
| 313 | 65532 | z = _mm256_sub_ps(z, _mm256_and_ps(_mm256_mul_ps(z, ftwos), condition)); | |
| 314 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_LT_OS); | |
| 315 | 98298 | x = _mm256_add_ps( | |
| 316 | z, _mm256_and_ps(_mm256_sub_ps(_mm256_div_ps(fones, z), z), condition)); | ||
| 317 | |||
| 318 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (i = 0; i < 2; i++) |
| 319 | 196596 | x = _mm256_add_ps(x, _mm256_sqrt_ps(_mm256_fmadd_ps(x, x, fones))); | |
| 320 | 32766 | x = _mm256_div_ps(fones, x); | |
| 321 | 32766 | y = fzeroes; | |
| 322 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (j = ACOS_TERMS - 1; j >= 0; j--) |
| 323 | 196596 | y = _mm256_fmadd_ps( | |
| 324 | 65532 | y, _mm256_mul_ps(x, x), _mm256_set1_ps(pow(-1, j) / (2 * j + 1))); | |
| 325 | |||
| 326 | 32766 | y = _mm256_mul_ps(y, _mm256_mul_ps(x, ffours)); | |
| 327 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_GT_OS); | |
| 328 | |||
| 329 | 65532 | y = _mm256_add_ps(y, _mm256_and_ps(_mm256_fnmadd_ps(y, ftwos, pio2), condition)); | |
| 330 | 32766 | arccosine = y; | |
| 331 | 32766 | condition = _mm256_cmp_ps(aVal, fzeroes, _CMP_LT_OS); | |
| 332 | 65532 | arccosine = _mm256_sub_ps( | |
| 333 | arccosine, _mm256_and_ps(_mm256_mul_ps(arccosine, ftwos), condition)); | ||
| 334 | 32766 | condition = _mm256_cmp_ps(d, fzeroes, _CMP_LT_OS); | |
| 335 | 65532 | arccosine = _mm256_add_ps(arccosine, _mm256_and_ps(pi, condition)); | |
| 336 | |||
| 337 | _mm256_storeu_ps(bPtr, arccosine); | ||
| 338 | 32766 | aPtr += 8; | |
| 339 | 32766 | bPtr += 8; | |
| 340 | } | ||
| 341 | |||
| 342 | 2 | number = eighthPoints * 8; | |
| 343 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 344 | 14 | *bPtr++ = acos(*aPtr++); | |
| 345 | } | ||
| 346 | 2 | } | |
| 347 | |||
| 348 | #endif /* LV_HAVE_AVX2 && LV_HAVE_FMA for unaligned */ | ||
| 349 | |||
| 350 | |||
| 351 | #ifdef LV_HAVE_AVX | ||
| 352 | #include <immintrin.h> | ||
| 353 | |||
| 354 | static inline void | ||
| 355 | 2 | volk_32f_acos_32f_u_avx(float* bVector, const float* aVector, unsigned int num_points) | |
| 356 | { | ||
| 357 | 2 | float* bPtr = bVector; | |
| 358 | 2 | const float* aPtr = aVector; | |
| 359 | |||
| 360 | 2 | unsigned int number = 0; | |
| 361 | 2 | unsigned int eighthPoints = num_points / 8; | |
| 362 | int i, j; | ||
| 363 | |||
| 364 | __m256 aVal, d, pi, pio2, x, y, z, arccosine; | ||
| 365 | __m256 fzeroes, fones, ftwos, ffours, condition; | ||
| 366 | |||
| 367 | 2 | pi = _mm256_set1_ps(3.14159265358979323846); | |
| 368 | 2 | pio2 = _mm256_set1_ps(3.14159265358979323846 / 2); | |
| 369 | 2 | fzeroes = _mm256_setzero_ps(); | |
| 370 | 2 | fones = _mm256_set1_ps(1.0); | |
| 371 | 2 | ftwos = _mm256_set1_ps(2.0); | |
| 372 | 2 | ffours = _mm256_set1_ps(4.0); | |
| 373 | |||
| 374 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 375 | 32766 | aVal = _mm256_loadu_ps(aPtr); | |
| 376 | 32766 | d = aVal; | |
| 377 | 131064 | aVal = _mm256_div_ps(_mm256_sqrt_ps(_mm256_mul_ps(_mm256_add_ps(fones, aVal), | |
| 378 | _mm256_sub_ps(fones, aVal))), | ||
| 379 | aVal); | ||
| 380 | 32766 | z = aVal; | |
| 381 | 32766 | condition = _mm256_cmp_ps(z, fzeroes, _CMP_LT_OS); | |
| 382 | 65532 | z = _mm256_sub_ps(z, _mm256_and_ps(_mm256_mul_ps(z, ftwos), condition)); | |
| 383 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_LT_OS); | |
| 384 | 98298 | x = _mm256_add_ps( | |
| 385 | z, _mm256_and_ps(_mm256_sub_ps(_mm256_div_ps(fones, z), z), condition)); | ||
| 386 | |||
| 387 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (i = 0; i < 2; i++) |
| 388 | 262128 | x = _mm256_add_ps(x, | |
| 389 | _mm256_sqrt_ps(_mm256_add_ps(fones, _mm256_mul_ps(x, x)))); | ||
| 390 | 32766 | x = _mm256_div_ps(fones, x); | |
| 391 | 32766 | y = fzeroes; | |
| 392 |
2/2✓ Branch 0 taken 65532 times.
✓ Branch 1 taken 32766 times.
|
98298 | for (j = ACOS_TERMS - 1; j >= 0; j--) |
| 393 | 262128 | y = _mm256_add_ps(_mm256_mul_ps(y, _mm256_mul_ps(x, x)), | |
| 394 | 65532 | _mm256_set1_ps(pow(-1, j) / (2 * j + 1))); | |
| 395 | |||
| 396 | 32766 | y = _mm256_mul_ps(y, _mm256_mul_ps(x, ffours)); | |
| 397 | 32766 | condition = _mm256_cmp_ps(z, fones, _CMP_GT_OS); | |
| 398 | |||
| 399 | 98298 | y = _mm256_add_ps( | |
| 400 | y, _mm256_and_ps(_mm256_sub_ps(pio2, _mm256_mul_ps(y, ftwos)), condition)); | ||
| 401 | 32766 | arccosine = y; | |
| 402 | 32766 | condition = _mm256_cmp_ps(aVal, fzeroes, _CMP_LT_OS); | |
| 403 | 65532 | arccosine = _mm256_sub_ps( | |
| 404 | arccosine, _mm256_and_ps(_mm256_mul_ps(arccosine, ftwos), condition)); | ||
| 405 | 32766 | condition = _mm256_cmp_ps(d, fzeroes, _CMP_LT_OS); | |
| 406 | 65532 | arccosine = _mm256_add_ps(arccosine, _mm256_and_ps(pi, condition)); | |
| 407 | |||
| 408 | _mm256_storeu_ps(bPtr, arccosine); | ||
| 409 | 32766 | aPtr += 8; | |
| 410 | 32766 | bPtr += 8; | |
| 411 | } | ||
| 412 | |||
| 413 | 2 | number = eighthPoints * 8; | |
| 414 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 415 | 14 | *bPtr++ = acos(*aPtr++); | |
| 416 | } | ||
| 417 | 2 | } | |
| 418 | |||
| 419 | #endif /* LV_HAVE_AVX2 for unaligned */ | ||
| 420 | |||
| 421 | #ifdef LV_HAVE_SSE4_1 | ||
| 422 | #include <smmintrin.h> | ||
| 423 | |||
| 424 | static inline void | ||
| 425 | 2 | volk_32f_acos_32f_u_sse4_1(float* bVector, const float* aVector, unsigned int num_points) | |
| 426 | { | ||
| 427 | 2 | float* bPtr = bVector; | |
| 428 | 2 | const float* aPtr = aVector; | |
| 429 | |||
| 430 | 2 | unsigned int number = 0; | |
| 431 | 2 | unsigned int quarterPoints = num_points / 4; | |
| 432 | int i, j; | ||
| 433 | |||
| 434 | __m128 aVal, d, pi, pio2, x, y, z, arccosine; | ||
| 435 | __m128 fzeroes, fones, ftwos, ffours, condition; | ||
| 436 | |||
| 437 | 2 | pi = _mm_set1_ps(3.14159265358979323846); | |
| 438 | 2 | pio2 = _mm_set1_ps(3.14159265358979323846 / 2); | |
| 439 | 2 | fzeroes = _mm_setzero_ps(); | |
| 440 | 2 | fones = _mm_set1_ps(1.0); | |
| 441 | 2 | ftwos = _mm_set1_ps(2.0); | |
| 442 | 2 | ffours = _mm_set1_ps(4.0); | |
| 443 | |||
| 444 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 445 | 65534 | aVal = _mm_loadu_ps(aPtr); | |
| 446 | 65534 | d = aVal; | |
| 447 | 262136 | aVal = _mm_div_ps( | |
| 448 | _mm_sqrt_ps(_mm_mul_ps(_mm_add_ps(fones, aVal), _mm_sub_ps(fones, aVal))), | ||
| 449 | aVal); | ||
| 450 | 65534 | z = aVal; | |
| 451 | 65534 | condition = _mm_cmplt_ps(z, fzeroes); | |
| 452 | 196602 | z = _mm_sub_ps(z, _mm_and_ps(_mm_mul_ps(z, ftwos), condition)); | |
| 453 | 65534 | condition = _mm_cmplt_ps(z, fones); | |
| 454 | 196602 | x = _mm_add_ps(z, _mm_and_ps(_mm_sub_ps(_mm_div_ps(fones, z), z), condition)); | |
| 455 | |||
| 456 |
2/2✓ Branch 0 taken 131068 times.
✓ Branch 1 taken 65534 times.
|
196602 | for (i = 0; i < 2; i++) |
| 457 | 524272 | x = _mm_add_ps(x, _mm_sqrt_ps(_mm_add_ps(fones, _mm_mul_ps(x, x)))); | |
| 458 | 65534 | x = _mm_div_ps(fones, x); | |
| 459 | 65534 | y = fzeroes; | |
| 460 | |||
| 461 |
2/2✓ Branch 0 taken 131068 times.
✓ Branch 1 taken 65534 times.
|
196602 | for (j = ACOS_TERMS - 1; j >= 0; j--) |
| 462 | 524272 | y = _mm_add_ps(_mm_mul_ps(y, _mm_mul_ps(x, x)), | |
| 463 | 131068 | _mm_set1_ps(pow(-1, j) / (2 * j + 1))); | |
| 464 | |||
| 465 | 131068 | y = _mm_mul_ps(y, _mm_mul_ps(x, ffours)); | |
| 466 | 65534 | condition = _mm_cmpgt_ps(z, fones); | |
| 467 | |||
| 468 | 196602 | y = _mm_add_ps(y, _mm_and_ps(_mm_sub_ps(pio2, _mm_mul_ps(y, ftwos)), condition)); | |
| 469 | 65534 | arccosine = y; | |
| 470 | 65534 | condition = _mm_cmplt_ps(aVal, fzeroes); | |
| 471 | arccosine = | ||
| 472 | 196602 | _mm_sub_ps(arccosine, _mm_and_ps(_mm_mul_ps(arccosine, ftwos), condition)); | |
| 473 | 65534 | condition = _mm_cmplt_ps(d, fzeroes); | |
| 474 | 131068 | arccosine = _mm_add_ps(arccosine, _mm_and_ps(pi, condition)); | |
| 475 | |||
| 476 | _mm_storeu_ps(bPtr, arccosine); | ||
| 477 | 65534 | aPtr += 4; | |
| 478 | 65534 | bPtr += 4; | |
| 479 | } | ||
| 480 | |||
| 481 | 2 | number = quarterPoints * 4; | |
| 482 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 483 | 6 | *bPtr++ = acosf(*aPtr++); | |
| 484 | } | ||
| 485 | 2 | } | |
| 486 | |||
| 487 | #endif /* LV_HAVE_SSE4_1 for aligned */ | ||
| 488 | |||
| 489 | #ifdef LV_HAVE_GENERIC | ||
| 490 | |||
| 491 | static inline void | ||
| 492 | 2 | volk_32f_acos_32f_generic(float* bVector, const float* aVector, unsigned int num_points) | |
| 493 | { | ||
| 494 | 2 | float* bPtr = bVector; | |
| 495 | 2 | const float* aPtr = aVector; | |
| 496 | 2 | unsigned int number = 0; | |
| 497 | |||
| 498 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
| 499 | 262142 | *bPtr++ = acosf(*aPtr++); | |
| 500 | } | ||
| 501 | 2 | } | |
| 502 | #endif /* LV_HAVE_GENERIC */ | ||
| 503 | |||
| 504 | #endif /* INCLUDED_volk_32f_acos_32f_u_H */ | ||
| 505 |