| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* -*- c++ -*- */ | ||
| 2 | /* | ||
| 3 | * Copyright 2012, 2013, 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_s32fc_x2_rotator_32fc | ||
| 12 | * | ||
| 13 | * \b Overview | ||
| 14 | * | ||
| 15 | * Rotate input vector at fixed rate per sample from initial phase | ||
| 16 | * offset. | ||
| 17 | * | ||
| 18 | * <b>Dispatcher Prototype</b> | ||
| 19 | * \code | ||
| 20 | * void volk_32fc_s32fc_x2_rotator_32fc(lv_32fc_t* outVector, const lv_32fc_t* inVector, | ||
| 21 | * const lv_32fc_t phase_inc, lv_32fc_t* phase, unsigned int num_points) \endcode | ||
| 22 | * | ||
| 23 | * \b Inputs | ||
| 24 | * \li inVector: Vector to be rotated. | ||
| 25 | * \li phase_inc: rotational velocity. | ||
| 26 | * \li phase: initial phase offset. | ||
| 27 | * \li num_points: The number of values in inVector to be rotated and stored into | ||
| 28 | * outVector. | ||
| 29 | * | ||
| 30 | * \b Outputs | ||
| 31 | * \li outVector: The vector where the results will be stored. | ||
| 32 | * | ||
| 33 | * \b Example | ||
| 34 | * Generate a tone at f=0.3 (normalized frequency) and use the rotator with | ||
| 35 | * f=0.1 to shift the tone to f=0.4. Change this example to start with a DC | ||
| 36 | * tone (initialize in with lv_cmake(1, 0)) to observe rotator signal generation. | ||
| 37 | * \code | ||
| 38 | * int N = 10; | ||
| 39 | * unsigned int alignment = volk_get_alignment(); | ||
| 40 | * lv_32fc_t* in = (lv_32fc_t*)volk_malloc(sizeof(lv_32fc_t)*N, alignment); | ||
| 41 | * lv_32fc_t* out = (lv_32fc_t*)volk_malloc(sizeof(lv_32fc_t)*N, alignment); | ||
| 42 | * | ||
| 43 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 44 | * // Generate a tone at f=0.3 | ||
| 45 | * float real = std::cos(0.3f * (float)ii); | ||
| 46 | * float imag = std::sin(0.3f * (float)ii); | ||
| 47 | * in[ii] = lv_cmake(real, imag); | ||
| 48 | * } | ||
| 49 | * // The oscillator rotates at f=0.1 | ||
| 50 | * float frequency = 0.1f; | ||
| 51 | * lv_32fc_t phase_increment = lv_cmake(std::cos(frequency), std::sin(frequency)); | ||
| 52 | * lv_32fc_t phase= lv_cmake(1.f, 0.0f); // start at 1 (0 rad phase) | ||
| 53 | * | ||
| 54 | * // rotate so the output is a tone at f=0.4 | ||
| 55 | * volk_32fc_s32fc_x2_rotator_32fc(out, in, phase_increment, &phase, N); | ||
| 56 | * | ||
| 57 | * // print results for inspection | ||
| 58 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 59 | * printf("out[%u] = %+1.2f %+1.2fj\n", | ||
| 60 | * ii, lv_creal(out[ii]), lv_cimag(out[ii])); | ||
| 61 | * } | ||
| 62 | * | ||
| 63 | * volk_free(in); | ||
| 64 | * volk_free(out); | ||
| 65 | * \endcode | ||
| 66 | */ | ||
| 67 | |||
| 68 | #ifndef INCLUDED_volk_32fc_s32fc_rotator_32fc_a_H | ||
| 69 | #define INCLUDED_volk_32fc_s32fc_rotator_32fc_a_H | ||
| 70 | |||
| 71 | |||
| 72 | #include <math.h> | ||
| 73 | #include <stdio.h> | ||
| 74 | #include <stdlib.h> | ||
| 75 | #include <volk/volk_complex.h> | ||
| 76 | #define ROTATOR_RELOAD 512 | ||
| 77 | #define ROTATOR_RELOAD_2 (ROTATOR_RELOAD / 2) | ||
| 78 | #define ROTATOR_RELOAD_4 (ROTATOR_RELOAD / 4) | ||
| 79 | |||
| 80 | |||
| 81 | #ifdef LV_HAVE_GENERIC | ||
| 82 | |||
| 83 | 6 | static inline void volk_32fc_s32fc_x2_rotator_32fc_generic(lv_32fc_t* outVector, | |
| 84 | const lv_32fc_t* inVector, | ||
| 85 | const lv_32fc_t phase_inc, | ||
| 86 | lv_32fc_t* phase, | ||
| 87 | unsigned int num_points) | ||
| 88 | { | ||
| 89 | 6 | unsigned int i = 0; | |
| 90 | 6 | int j = 0; | |
| 91 |
2/2✓ Branch 0 taken 510 times.
✓ Branch 1 taken 6 times.
|
516 | for (i = 0; i < (unsigned int)(num_points / ROTATOR_RELOAD); ++i) { |
| 92 |
2/2✓ Branch 0 taken 261120 times.
✓ Branch 1 taken 510 times.
|
261630 | for (j = 0; j < ROTATOR_RELOAD; ++j) { |
| 93 | 261120 | *outVector++ = *inVector++ * (*phase); | |
| 94 | 261120 | (*phase) *= phase_inc; | |
| 95 | } | ||
| 96 | |||
| 97 | 510 | (*phase) /= hypotf(lv_creal(*phase), lv_cimag(*phase)); | |
| 98 | } | ||
| 99 |
2/2✓ Branch 0 taken 1034 times.
✓ Branch 1 taken 6 times.
|
1040 | for (i = 0; i < num_points % ROTATOR_RELOAD; ++i) { |
| 100 | 1034 | *outVector++ = *inVector++ * (*phase); | |
| 101 | 1034 | (*phase) *= phase_inc; | |
| 102 | } | ||
| 103 |
1/2✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
|
6 | if (i) { |
| 104 | // Make sure, we normalize phase on every call! | ||
| 105 | 6 | (*phase) /= hypotf(lv_creal(*phase), lv_cimag(*phase)); | |
| 106 | } | ||
| 107 | 6 | } | |
| 108 | |||
| 109 | #endif /* LV_HAVE_GENERIC */ | ||
| 110 | |||
| 111 | |||
| 112 | #ifdef LV_HAVE_NEON | ||
| 113 | #include <arm_neon.h> | ||
| 114 | #include <volk/volk_neon_intrinsics.h> | ||
| 115 | |||
| 116 | static inline void volk_32fc_s32fc_x2_rotator_32fc_neon(lv_32fc_t* outVector, | ||
| 117 | const lv_32fc_t* inVector, | ||
| 118 | const lv_32fc_t phase_inc, | ||
| 119 | lv_32fc_t* phase, | ||
| 120 | unsigned int num_points) | ||
| 121 | |||
| 122 | { | ||
| 123 | lv_32fc_t* outputVectorPtr = outVector; | ||
| 124 | const lv_32fc_t* inputVectorPtr = inVector; | ||
| 125 | lv_32fc_t incr = 1; | ||
| 126 | lv_32fc_t phasePtr[4] = { (*phase), (*phase), (*phase), (*phase) }; | ||
| 127 | float32x4x2_t input_vec; | ||
| 128 | float32x4x2_t output_vec; | ||
| 129 | |||
| 130 | unsigned int i = 0, j = 0; | ||
| 131 | // const unsigned int quarter_points = num_points / 4; | ||
| 132 | |||
| 133 | for (i = 0; i < 4; ++i) { | ||
| 134 | phasePtr[i] *= incr; | ||
| 135 | incr *= (phase_inc); | ||
| 136 | } | ||
| 137 | |||
| 138 | // Notice that incr has be incremented in the previous loop | ||
| 139 | const lv_32fc_t incrPtr[4] = { incr, incr, incr, incr }; | ||
| 140 | const float32x4x2_t incr_vec = vld2q_f32((float*)incrPtr); | ||
| 141 | float32x4x2_t phase_vec = vld2q_f32((float*)phasePtr); | ||
| 142 | |||
| 143 | for (i = 0; i < (unsigned int)(num_points / ROTATOR_RELOAD); i++) { | ||
| 144 | for (j = 0; j < ROTATOR_RELOAD_4; j++) { | ||
| 145 | input_vec = vld2q_f32((float*)inputVectorPtr); | ||
| 146 | // Prefetch next one, speeds things up | ||
| 147 | __VOLK_PREFETCH(inputVectorPtr + 4); | ||
| 148 | // Rotate | ||
| 149 | output_vec = _vmultiply_complexq_f32(input_vec, phase_vec); | ||
| 150 | // Increase phase | ||
| 151 | phase_vec = _vmultiply_complexq_f32(phase_vec, incr_vec); | ||
| 152 | // Store output | ||
| 153 | vst2q_f32((float*)outputVectorPtr, output_vec); | ||
| 154 | |||
| 155 | outputVectorPtr += 4; | ||
| 156 | inputVectorPtr += 4; | ||
| 157 | } | ||
| 158 | // normalize phase so magnitude doesn't grow because of | ||
| 159 | // floating point rounding error | ||
| 160 | const float32x4_t mag_squared = _vmagnitudesquaredq_f32(phase_vec); | ||
| 161 | const float32x4_t inv_mag = _vinvsqrtq_f32(mag_squared); | ||
| 162 | // Multiply complex with real | ||
| 163 | phase_vec.val[0] = vmulq_f32(phase_vec.val[0], inv_mag); | ||
| 164 | phase_vec.val[1] = vmulq_f32(phase_vec.val[1], inv_mag); | ||
| 165 | } | ||
| 166 | |||
| 167 | for (i = 0; i < (num_points % ROTATOR_RELOAD) / 4; i++) { | ||
| 168 | input_vec = vld2q_f32((float*)inputVectorPtr); | ||
| 169 | // Prefetch next one, speeds things up | ||
| 170 | __VOLK_PREFETCH(inputVectorPtr + 4); | ||
| 171 | // Rotate | ||
| 172 | output_vec = _vmultiply_complexq_f32(input_vec, phase_vec); | ||
| 173 | // Increase phase | ||
| 174 | phase_vec = _vmultiply_complexq_f32(phase_vec, incr_vec); | ||
| 175 | // Store output | ||
| 176 | vst2q_f32((float*)outputVectorPtr, output_vec); | ||
| 177 | |||
| 178 | outputVectorPtr += 4; | ||
| 179 | inputVectorPtr += 4; | ||
| 180 | } | ||
| 181 | // if(i) == true means we looped above | ||
| 182 | if (i) { | ||
| 183 | // normalize phase so magnitude doesn't grow because of | ||
| 184 | // floating point rounding error | ||
| 185 | const float32x4_t mag_squared = _vmagnitudesquaredq_f32(phase_vec); | ||
| 186 | const float32x4_t inv_mag = _vinvsqrtq_f32(mag_squared); | ||
| 187 | // Multiply complex with real | ||
| 188 | phase_vec.val[0] = vmulq_f32(phase_vec.val[0], inv_mag); | ||
| 189 | phase_vec.val[1] = vmulq_f32(phase_vec.val[1], inv_mag); | ||
| 190 | } | ||
| 191 | // Store current phase | ||
| 192 | vst2q_f32((float*)phasePtr, phase_vec); | ||
| 193 | |||
| 194 | // Deal with the rest | ||
| 195 | for (i = 0; i < num_points % 4; i++) { | ||
| 196 | *outputVectorPtr++ = *inputVectorPtr++ * phasePtr[0]; | ||
| 197 | phasePtr[0] *= (phase_inc); | ||
| 198 | } | ||
| 199 | |||
| 200 | // For continuous phase next time we need to call this function | ||
| 201 | (*phase) = phasePtr[0]; | ||
| 202 | } | ||
| 203 | |||
| 204 | #endif /* LV_HAVE_NEON */ | ||
| 205 | |||
| 206 | |||
| 207 | #ifdef LV_HAVE_SSE4_1 | ||
| 208 | #include <smmintrin.h> | ||
| 209 | |||
| 210 | 2 | static inline void volk_32fc_s32fc_x2_rotator_32fc_a_sse4_1(lv_32fc_t* outVector, | |
| 211 | const lv_32fc_t* inVector, | ||
| 212 | const lv_32fc_t phase_inc, | ||
| 213 | lv_32fc_t* phase, | ||
| 214 | unsigned int num_points) | ||
| 215 | { | ||
| 216 | 2 | lv_32fc_t* cPtr = outVector; | |
| 217 | 2 | const lv_32fc_t* aPtr = inVector; | |
| 218 | 2 | lv_32fc_t incr = 1; | |
| 219 | 2 | lv_32fc_t phase_Ptr[2] = { (*phase), (*phase) }; | |
| 220 | |||
| 221 | 2 | unsigned int i, j = 0; | |
| 222 | |||
| 223 |
2/2✓ Branch 0 taken 4 times.
✓ Branch 1 taken 2 times.
|
6 | for (i = 0; i < 2; ++i) { |
| 224 | 4 | phase_Ptr[i] *= incr; | |
| 225 | 4 | incr *= (phase_inc); | |
| 226 | } | ||
| 227 | |||
| 228 | __m128 aVal, phase_Val, inc_Val, yl, yh, tmp1, tmp2, z, ylp, yhp, tmp1p, tmp2p; | ||
| 229 | |||
| 230 | 2 | phase_Val = _mm_loadu_ps((float*)phase_Ptr); | |
| 231 | 2 | inc_Val = _mm_set_ps(lv_cimag(incr), lv_creal(incr), lv_cimag(incr), lv_creal(incr)); | |
| 232 | |||
| 233 |
2/2✓ Branch 0 taken 510 times.
✓ Branch 1 taken 2 times.
|
512 | for (i = 0; i < (unsigned int)(num_points / ROTATOR_RELOAD); i++) { |
| 234 |
2/2✓ Branch 0 taken 130560 times.
✓ Branch 1 taken 510 times.
|
131070 | for (j = 0; j < ROTATOR_RELOAD_2; ++j) { |
| 235 | |||
| 236 | 130560 | aVal = _mm_load_ps((float*)aPtr); | |
| 237 | |||
| 238 | 130560 | yl = _mm_moveldup_ps(phase_Val); | |
| 239 | 130560 | yh = _mm_movehdup_ps(phase_Val); | |
| 240 | 130560 | ylp = _mm_moveldup_ps(inc_Val); | |
| 241 | 130560 | yhp = _mm_movehdup_ps(inc_Val); | |
| 242 | |||
| 243 | 130560 | tmp1 = _mm_mul_ps(aVal, yl); | |
| 244 | 130560 | tmp1p = _mm_mul_ps(phase_Val, ylp); | |
| 245 | |||
| 246 | 130560 | aVal = _mm_shuffle_ps(aVal, aVal, 0xB1); | |
| 247 | 130560 | phase_Val = _mm_shuffle_ps(phase_Val, phase_Val, 0xB1); | |
| 248 | 130560 | tmp2 = _mm_mul_ps(aVal, yh); | |
| 249 | 130560 | tmp2p = _mm_mul_ps(phase_Val, yhp); | |
| 250 | |||
| 251 | 130560 | z = _mm_addsub_ps(tmp1, tmp2); | |
| 252 | 130560 | phase_Val = _mm_addsub_ps(tmp1p, tmp2p); | |
| 253 | |||
| 254 | _mm_store_ps((float*)cPtr, z); | ||
| 255 | |||
| 256 | 130560 | aPtr += 2; | |
| 257 | 130560 | cPtr += 2; | |
| 258 | } | ||
| 259 | 510 | tmp1 = _mm_mul_ps(phase_Val, phase_Val); | |
| 260 | 510 | tmp2 = _mm_hadd_ps(tmp1, tmp1); | |
| 261 | 510 | tmp1 = _mm_shuffle_ps(tmp2, tmp2, 0xD8); | |
| 262 | 510 | tmp2 = _mm_sqrt_ps(tmp1); | |
| 263 | 510 | phase_Val = _mm_div_ps(phase_Val, tmp2); | |
| 264 | } | ||
| 265 |
2/2✓ Branch 0 taken 510 times.
✓ Branch 1 taken 2 times.
|
512 | for (i = 0; i < (num_points % ROTATOR_RELOAD) / 2; ++i) { |
| 266 | 510 | aVal = _mm_load_ps((float*)aPtr); | |
| 267 | |||
| 268 | 510 | yl = _mm_moveldup_ps(phase_Val); | |
| 269 | 510 | yh = _mm_movehdup_ps(phase_Val); | |
| 270 | 510 | ylp = _mm_moveldup_ps(inc_Val); | |
| 271 | 510 | yhp = _mm_movehdup_ps(inc_Val); | |
| 272 | |||
| 273 | 510 | tmp1 = _mm_mul_ps(aVal, yl); | |
| 274 | |||
| 275 | 510 | tmp1p = _mm_mul_ps(phase_Val, ylp); | |
| 276 | |||
| 277 | 510 | aVal = _mm_shuffle_ps(aVal, aVal, 0xB1); | |
| 278 | 510 | phase_Val = _mm_shuffle_ps(phase_Val, phase_Val, 0xB1); | |
| 279 | 510 | tmp2 = _mm_mul_ps(aVal, yh); | |
| 280 | 510 | tmp2p = _mm_mul_ps(phase_Val, yhp); | |
| 281 | |||
| 282 | 510 | z = _mm_addsub_ps(tmp1, tmp2); | |
| 283 | 510 | phase_Val = _mm_addsub_ps(tmp1p, tmp2p); | |
| 284 | |||
| 285 | _mm_store_ps((float*)cPtr, z); | ||
| 286 | |||
| 287 | 510 | aPtr += 2; | |
| 288 | 510 | cPtr += 2; | |
| 289 | } | ||
| 290 |
1/2✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
|
2 | if (i) { |
| 291 | 2 | tmp1 = _mm_mul_ps(phase_Val, phase_Val); | |
| 292 | 2 | tmp2 = _mm_hadd_ps(tmp1, tmp1); | |
| 293 | 2 | tmp1 = _mm_shuffle_ps(tmp2, tmp2, 0xD8); | |
| 294 | 2 | tmp2 = _mm_sqrt_ps(tmp1); | |
| 295 | 2 | phase_Val = _mm_div_ps(phase_Val, tmp2); | |
| 296 | } | ||
| 297 | |||
| 298 | _mm_storeu_ps((float*)phase_Ptr, phase_Val); | ||
| 299 |
1/2✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
|
2 | if (num_points & 1) { |
| 300 | 2 | *cPtr++ = *aPtr++ * phase_Ptr[0]; | |
| 301 | 2 | phase_Ptr[0] *= (phase_inc); | |
| 302 | } | ||
| 303 | |||
| 304 | 2 | (*phase) = phase_Ptr[0]; | |
| 305 | 2 | } | |
| 306 | |||
| 307 | #endif /* LV_HAVE_SSE4_1 for aligned */ | ||
| 308 | |||
| 309 | |||
| 310 | #ifdef LV_HAVE_SSE4_1 | ||
| 311 | #include <smmintrin.h> | ||
| 312 | |||
| 313 | 2 | static inline void volk_32fc_s32fc_x2_rotator_32fc_u_sse4_1(lv_32fc_t* outVector, | |
| 314 | const lv_32fc_t* inVector, | ||
| 315 | const lv_32fc_t phase_inc, | ||
| 316 | lv_32fc_t* phase, | ||
| 317 | unsigned int num_points) | ||
| 318 | { | ||
| 319 | 2 | lv_32fc_t* cPtr = outVector; | |
| 320 | 2 | const lv_32fc_t* aPtr = inVector; | |
| 321 | 2 | lv_32fc_t incr = 1; | |
| 322 | 2 | lv_32fc_t phase_Ptr[2] = { (*phase), (*phase) }; | |
| 323 | |||
| 324 | 2 | unsigned int i, j = 0; | |
| 325 | |||
| 326 |
2/2✓ Branch 0 taken 4 times.
✓ Branch 1 taken 2 times.
|
6 | for (i = 0; i < 2; ++i) { |
| 327 | 4 | phase_Ptr[i] *= incr; | |
| 328 | 4 | incr *= (phase_inc); | |
| 329 | } | ||
| 330 | |||
| 331 | /*printf("%f, %f\n", lv_creal(phase_Ptr[0]), lv_cimag(phase_Ptr[0])); | ||
| 332 | printf("%f, %f\n", lv_creal(phase_Ptr[1]), lv_cimag(phase_Ptr[1])); | ||
| 333 | printf("incr: %f, %f\n", lv_creal(incr), lv_cimag(incr));*/ | ||
| 334 | __m128 aVal, phase_Val, inc_Val, yl, yh, tmp1, tmp2, z, ylp, yhp, tmp1p, tmp2p; | ||
| 335 | |||
| 336 | 2 | phase_Val = _mm_loadu_ps((float*)phase_Ptr); | |
| 337 | 2 | inc_Val = _mm_set_ps(lv_cimag(incr), lv_creal(incr), lv_cimag(incr), lv_creal(incr)); | |
| 338 | |||
| 339 |
2/2✓ Branch 0 taken 510 times.
✓ Branch 1 taken 2 times.
|
512 | for (i = 0; i < (unsigned int)(num_points / ROTATOR_RELOAD); i++) { |
| 340 |
2/2✓ Branch 0 taken 130560 times.
✓ Branch 1 taken 510 times.
|
131070 | for (j = 0; j < ROTATOR_RELOAD_2; ++j) { |
| 341 | |||
| 342 | 130560 | aVal = _mm_loadu_ps((float*)aPtr); | |
| 343 | |||
| 344 | 130560 | yl = _mm_moveldup_ps(phase_Val); | |
| 345 | 130560 | yh = _mm_movehdup_ps(phase_Val); | |
| 346 | 130560 | ylp = _mm_moveldup_ps(inc_Val); | |
| 347 | 130560 | yhp = _mm_movehdup_ps(inc_Val); | |
| 348 | |||
| 349 | 130560 | tmp1 = _mm_mul_ps(aVal, yl); | |
| 350 | 130560 | tmp1p = _mm_mul_ps(phase_Val, ylp); | |
| 351 | |||
| 352 | 130560 | aVal = _mm_shuffle_ps(aVal, aVal, 0xB1); | |
| 353 | 130560 | phase_Val = _mm_shuffle_ps(phase_Val, phase_Val, 0xB1); | |
| 354 | 130560 | tmp2 = _mm_mul_ps(aVal, yh); | |
| 355 | 130560 | tmp2p = _mm_mul_ps(phase_Val, yhp); | |
| 356 | |||
| 357 | 130560 | z = _mm_addsub_ps(tmp1, tmp2); | |
| 358 | 130560 | phase_Val = _mm_addsub_ps(tmp1p, tmp2p); | |
| 359 | |||
| 360 | _mm_storeu_ps((float*)cPtr, z); | ||
| 361 | |||
| 362 | 130560 | aPtr += 2; | |
| 363 | 130560 | cPtr += 2; | |
| 364 | } | ||
| 365 | 510 | tmp1 = _mm_mul_ps(phase_Val, phase_Val); | |
| 366 | 510 | tmp2 = _mm_hadd_ps(tmp1, tmp1); | |
| 367 | 510 | tmp1 = _mm_shuffle_ps(tmp2, tmp2, 0xD8); | |
| 368 | 510 | tmp2 = _mm_sqrt_ps(tmp1); | |
| 369 | 510 | phase_Val = _mm_div_ps(phase_Val, tmp2); | |
| 370 | } | ||
| 371 |
2/2✓ Branch 0 taken 510 times.
✓ Branch 1 taken 2 times.
|
512 | for (i = 0; i < (num_points % ROTATOR_RELOAD) / 2; ++i) { |
| 372 | 510 | aVal = _mm_loadu_ps((float*)aPtr); | |
| 373 | |||
| 374 | 510 | yl = _mm_moveldup_ps(phase_Val); | |
| 375 | 510 | yh = _mm_movehdup_ps(phase_Val); | |
| 376 | 510 | ylp = _mm_moveldup_ps(inc_Val); | |
| 377 | 510 | yhp = _mm_movehdup_ps(inc_Val); | |
| 378 | |||
| 379 | 510 | tmp1 = _mm_mul_ps(aVal, yl); | |
| 380 | |||
| 381 | 510 | tmp1p = _mm_mul_ps(phase_Val, ylp); | |
| 382 | |||
| 383 | 510 | aVal = _mm_shuffle_ps(aVal, aVal, 0xB1); | |
| 384 | 510 | phase_Val = _mm_shuffle_ps(phase_Val, phase_Val, 0xB1); | |
| 385 | 510 | tmp2 = _mm_mul_ps(aVal, yh); | |
| 386 | 510 | tmp2p = _mm_mul_ps(phase_Val, yhp); | |
| 387 | |||
| 388 | 510 | z = _mm_addsub_ps(tmp1, tmp2); | |
| 389 | 510 | phase_Val = _mm_addsub_ps(tmp1p, tmp2p); | |
| 390 | |||
| 391 | _mm_storeu_ps((float*)cPtr, z); | ||
| 392 | |||
| 393 | 510 | aPtr += 2; | |
| 394 | 510 | cPtr += 2; | |
| 395 | } | ||
| 396 |
1/2✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
|
2 | if (i) { |
| 397 | 2 | tmp1 = _mm_mul_ps(phase_Val, phase_Val); | |
| 398 | 2 | tmp2 = _mm_hadd_ps(tmp1, tmp1); | |
| 399 | 2 | tmp1 = _mm_shuffle_ps(tmp2, tmp2, 0xD8); | |
| 400 | 2 | tmp2 = _mm_sqrt_ps(tmp1); | |
| 401 | 2 | phase_Val = _mm_div_ps(phase_Val, tmp2); | |
| 402 | } | ||
| 403 | |||
| 404 | _mm_storeu_ps((float*)phase_Ptr, phase_Val); | ||
| 405 |
1/2✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
|
2 | if (num_points & 1) { |
| 406 | 2 | *cPtr++ = *aPtr++ * phase_Ptr[0]; | |
| 407 | 2 | phase_Ptr[0] *= (phase_inc); | |
| 408 | } | ||
| 409 | |||
| 410 | 2 | (*phase) = phase_Ptr[0]; | |
| 411 | 2 | } | |
| 412 | |||
| 413 | #endif /* LV_HAVE_SSE4_1 */ | ||
| 414 | |||
| 415 | |||
| 416 | #ifdef LV_HAVE_AVX | ||
| 417 | #include <immintrin.h> | ||
| 418 | #include <volk/volk_avx_intrinsics.h> | ||
| 419 | |||
| 420 | 2 | static inline void volk_32fc_s32fc_x2_rotator_32fc_a_avx(lv_32fc_t* outVector, | |
| 421 | const lv_32fc_t* inVector, | ||
| 422 | const lv_32fc_t phase_inc, | ||
| 423 | lv_32fc_t* phase, | ||
| 424 | unsigned int num_points) | ||
| 425 | { | ||
| 426 | 2 | lv_32fc_t* cPtr = outVector; | |
| 427 | 2 | const lv_32fc_t* aPtr = inVector; | |
| 428 | 2 | lv_32fc_t incr = lv_cmake(1.0f, 0.0f); | |
| 429 | 2 | lv_32fc_t phase_Ptr[4] = { (*phase), (*phase), (*phase), (*phase) }; | |
| 430 | |||
| 431 | 2 | unsigned int i, j = 0; | |
| 432 | |||
| 433 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 2 times.
|
10 | for (i = 0; i < 4; ++i) { |
| 434 | 8 | phase_Ptr[i] *= incr; | |
| 435 | 8 | incr *= (phase_inc); | |
| 436 | } | ||
| 437 | |||
| 438 | __m256 aVal, phase_Val, z; | ||
| 439 | |||
| 440 | 2 | phase_Val = _mm256_loadu_ps((float*)phase_Ptr); | |
| 441 | |||
| 442 | 2 | const __m256 inc_Val = _mm256_set_ps(lv_cimag(incr), | |
| 443 | lv_creal(incr), | ||
| 444 | lv_cimag(incr), | ||
| 445 | lv_creal(incr), | ||
| 446 | lv_cimag(incr), | ||
| 447 | lv_creal(incr), | ||
| 448 | lv_cimag(incr), | ||
| 449 | lv_creal(incr)); | ||
| 450 | |||
| 451 |
2/2✓ Branch 0 taken 510 times.
✓ Branch 1 taken 2 times.
|
512 | for (i = 0; i < (unsigned int)(num_points / ROTATOR_RELOAD); i++) { |
| 452 |
2/2✓ Branch 0 taken 65280 times.
✓ Branch 1 taken 510 times.
|
65790 | for (j = 0; j < ROTATOR_RELOAD_4; ++j) { |
| 453 | |||
| 454 | 65280 | aVal = _mm256_load_ps((float*)aPtr); | |
| 455 | |||
| 456 | 65280 | z = _mm256_complexmul_ps(aVal, phase_Val); | |
| 457 | 65280 | phase_Val = _mm256_complexmul_ps(phase_Val, inc_Val); | |
| 458 | |||
| 459 | _mm256_store_ps((float*)cPtr, z); | ||
| 460 | |||
| 461 | 65280 | aPtr += 4; | |
| 462 | 65280 | cPtr += 4; | |
| 463 | } | ||
| 464 | 510 | phase_Val = _mm256_normalize_ps(phase_Val); | |
| 465 | } | ||
| 466 | |||
| 467 |
2/2✓ Branch 0 taken 254 times.
✓ Branch 1 taken 2 times.
|
256 | for (i = 0; i < (num_points % ROTATOR_RELOAD) / 4; ++i) { |
| 468 | 254 | aVal = _mm256_load_ps((float*)aPtr); | |
| 469 | |||
| 470 | 254 | z = _mm256_complexmul_ps(aVal, phase_Val); | |
| 471 | 254 | phase_Val = _mm256_complexmul_ps(phase_Val, inc_Val); | |
| 472 | |||
| 473 | _mm256_store_ps((float*)cPtr, z); | ||
| 474 | |||
| 475 | 254 | aPtr += 4; | |
| 476 | 254 | cPtr += 4; | |
| 477 | } | ||
| 478 |
1/2✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
|
2 | if (i) { |
| 479 | 2 | phase_Val = _mm256_normalize_ps(phase_Val); | |
| 480 | } | ||
| 481 | |||
| 482 | _mm256_storeu_ps((float*)phase_Ptr, phase_Val); | ||
| 483 | 2 | (*phase) = phase_Ptr[0]; | |
| 484 | 2 | volk_32fc_s32fc_x2_rotator_32fc_generic(cPtr, aPtr, phase_inc, phase, num_points % 4); | |
| 485 | 2 | } | |
| 486 | |||
| 487 | #endif /* LV_HAVE_AVX for aligned */ | ||
| 488 | |||
| 489 | |||
| 490 | #ifdef LV_HAVE_AVX | ||
| 491 | #include <immintrin.h> | ||
| 492 | #include <volk/volk_avx_intrinsics.h> | ||
| 493 | |||
| 494 | 2 | static inline void volk_32fc_s32fc_x2_rotator_32fc_u_avx(lv_32fc_t* outVector, | |
| 495 | const lv_32fc_t* inVector, | ||
| 496 | const lv_32fc_t phase_inc, | ||
| 497 | lv_32fc_t* phase, | ||
| 498 | unsigned int num_points) | ||
| 499 | { | ||
| 500 | 2 | lv_32fc_t* cPtr = outVector; | |
| 501 | 2 | const lv_32fc_t* aPtr = inVector; | |
| 502 | 2 | lv_32fc_t incr = lv_cmake(1.0f, 0.0f); | |
| 503 | 2 | lv_32fc_t phase_Ptr[4] = { (*phase), (*phase), (*phase), (*phase) }; | |
| 504 | |||
| 505 | 2 | unsigned int i, j = 0; | |
| 506 | |||
| 507 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 2 times.
|
10 | for (i = 0; i < 4; ++i) { |
| 508 | 8 | phase_Ptr[i] *= incr; | |
| 509 | 8 | incr *= (phase_inc); | |
| 510 | } | ||
| 511 | |||
| 512 | __m256 aVal, phase_Val, z; | ||
| 513 | |||
| 514 | 2 | phase_Val = _mm256_loadu_ps((float*)phase_Ptr); | |
| 515 | |||
| 516 | 2 | const __m256 inc_Val = _mm256_set_ps(lv_cimag(incr), | |
| 517 | lv_creal(incr), | ||
| 518 | lv_cimag(incr), | ||
| 519 | lv_creal(incr), | ||
| 520 | lv_cimag(incr), | ||
| 521 | lv_creal(incr), | ||
| 522 | lv_cimag(incr), | ||
| 523 | lv_creal(incr)); | ||
| 524 | |||
| 525 |
2/2✓ Branch 0 taken 510 times.
✓ Branch 1 taken 2 times.
|
512 | for (i = 0; i < (unsigned int)(num_points / ROTATOR_RELOAD); ++i) { |
| 526 |
2/2✓ Branch 0 taken 65280 times.
✓ Branch 1 taken 510 times.
|
65790 | for (j = 0; j < ROTATOR_RELOAD_4; ++j) { |
| 527 | |||
| 528 | 65280 | aVal = _mm256_loadu_ps((float*)aPtr); | |
| 529 | |||
| 530 | 65280 | z = _mm256_complexmul_ps(aVal, phase_Val); | |
| 531 | 65280 | phase_Val = _mm256_complexmul_ps(phase_Val, inc_Val); | |
| 532 | |||
| 533 | _mm256_storeu_ps((float*)cPtr, z); | ||
| 534 | |||
| 535 | 65280 | aPtr += 4; | |
| 536 | 65280 | cPtr += 4; | |
| 537 | } | ||
| 538 | 510 | phase_Val = _mm256_normalize_ps(phase_Val); | |
| 539 | } | ||
| 540 | |||
| 541 |
2/2✓ Branch 0 taken 254 times.
✓ Branch 1 taken 2 times.
|
256 | for (i = 0; i < (num_points % ROTATOR_RELOAD) / 4; ++i) { |
| 542 | 254 | aVal = _mm256_loadu_ps((float*)aPtr); | |
| 543 | |||
| 544 | 254 | z = _mm256_complexmul_ps(aVal, phase_Val); | |
| 545 | 254 | phase_Val = _mm256_complexmul_ps(phase_Val, inc_Val); | |
| 546 | |||
| 547 | _mm256_storeu_ps((float*)cPtr, z); | ||
| 548 | |||
| 549 | 254 | aPtr += 4; | |
| 550 | 254 | cPtr += 4; | |
| 551 | } | ||
| 552 |
1/2✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
|
2 | if (i) { |
| 553 | 2 | phase_Val = _mm256_normalize_ps(phase_Val); | |
| 554 | } | ||
| 555 | |||
| 556 | _mm256_storeu_ps((float*)phase_Ptr, phase_Val); | ||
| 557 | 2 | (*phase) = phase_Ptr[0]; | |
| 558 | 2 | volk_32fc_s32fc_x2_rotator_32fc_generic(cPtr, aPtr, phase_inc, phase, num_points % 4); | |
| 559 | 2 | } | |
| 560 | |||
| 561 | #endif /* LV_HAVE_AVX */ | ||
| 562 | |||
| 563 | #if LV_HAVE_AVX && LV_HAVE_FMA | ||
| 564 | #include <immintrin.h> | ||
| 565 | |||
| 566 | 2 | static inline void volk_32fc_s32fc_x2_rotator_32fc_a_avx_fma(lv_32fc_t* outVector, | |
| 567 | const lv_32fc_t* inVector, | ||
| 568 | const lv_32fc_t phase_inc, | ||
| 569 | lv_32fc_t* phase, | ||
| 570 | unsigned int num_points) | ||
| 571 | { | ||
| 572 | 2 | lv_32fc_t* cPtr = outVector; | |
| 573 | 2 | const lv_32fc_t* aPtr = inVector; | |
| 574 | 2 | lv_32fc_t incr = 1; | |
| 575 | __VOLK_ATTR_ALIGNED(32) | ||
| 576 | 2 | lv_32fc_t phase_Ptr[4] = { (*phase), (*phase), (*phase), (*phase) }; | |
| 577 | |||
| 578 | 2 | unsigned int i, j = 0; | |
| 579 | |||
| 580 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 2 times.
|
10 | for (i = 0; i < 4; ++i) { |
| 581 | 8 | phase_Ptr[i] *= incr; | |
| 582 | 8 | incr *= (phase_inc); | |
| 583 | } | ||
| 584 | |||
| 585 | __m256 aVal, phase_Val, inc_Val, yl, yh, tmp1, tmp2, z, ylp, yhp, tmp1p, tmp2p; | ||
| 586 | |||
| 587 | 2 | phase_Val = _mm256_load_ps((float*)phase_Ptr); | |
| 588 | 2 | inc_Val = _mm256_set_ps(lv_cimag(incr), | |
| 589 | lv_creal(incr), | ||
| 590 | lv_cimag(incr), | ||
| 591 | lv_creal(incr), | ||
| 592 | lv_cimag(incr), | ||
| 593 | lv_creal(incr), | ||
| 594 | lv_cimag(incr), | ||
| 595 | lv_creal(incr)); | ||
| 596 | |||
| 597 |
2/2✓ Branch 0 taken 510 times.
✓ Branch 1 taken 2 times.
|
512 | for (i = 0; i < (unsigned int)(num_points / ROTATOR_RELOAD); i++) { |
| 598 |
2/2✓ Branch 0 taken 65280 times.
✓ Branch 1 taken 510 times.
|
65790 | for (j = 0; j < ROTATOR_RELOAD_4; ++j) { |
| 599 | |||
| 600 | 65280 | aVal = _mm256_load_ps((float*)aPtr); | |
| 601 | |||
| 602 | 65280 | yl = _mm256_moveldup_ps(phase_Val); | |
| 603 | 65280 | yh = _mm256_movehdup_ps(phase_Val); | |
| 604 | 65280 | ylp = _mm256_moveldup_ps(inc_Val); | |
| 605 | 65280 | yhp = _mm256_movehdup_ps(inc_Val); | |
| 606 | |||
| 607 | 65280 | tmp1 = aVal; | |
| 608 | 65280 | tmp1p = phase_Val; | |
| 609 | |||
| 610 | 65280 | aVal = _mm256_shuffle_ps(aVal, aVal, 0xB1); | |
| 611 | 65280 | phase_Val = _mm256_shuffle_ps(phase_Val, phase_Val, 0xB1); | |
| 612 | 65280 | tmp2 = _mm256_mul_ps(aVal, yh); | |
| 613 | 65280 | tmp2p = _mm256_mul_ps(phase_Val, yhp); | |
| 614 | |||
| 615 | 65280 | z = _mm256_fmaddsub_ps(tmp1, yl, tmp2); | |
| 616 | 65280 | phase_Val = _mm256_fmaddsub_ps(tmp1p, ylp, tmp2p); | |
| 617 | |||
| 618 | _mm256_store_ps((float*)cPtr, z); | ||
| 619 | |||
| 620 | 65280 | aPtr += 4; | |
| 621 | 65280 | cPtr += 4; | |
| 622 | } | ||
| 623 | 510 | tmp1 = _mm256_mul_ps(phase_Val, phase_Val); | |
| 624 | 510 | tmp2 = _mm256_hadd_ps(tmp1, tmp1); | |
| 625 | 510 | tmp1 = _mm256_shuffle_ps(tmp2, tmp2, 0xD8); | |
| 626 | 510 | tmp2 = _mm256_sqrt_ps(tmp1); | |
| 627 | 510 | phase_Val = _mm256_div_ps(phase_Val, tmp2); | |
| 628 | } | ||
| 629 |
2/2✓ Branch 0 taken 254 times.
✓ Branch 1 taken 2 times.
|
256 | for (i = 0; i < (num_points % ROTATOR_RELOAD) / 4; ++i) { |
| 630 | 254 | aVal = _mm256_load_ps((float*)aPtr); | |
| 631 | |||
| 632 | 254 | yl = _mm256_moveldup_ps(phase_Val); | |
| 633 | 254 | yh = _mm256_movehdup_ps(phase_Val); | |
| 634 | 254 | ylp = _mm256_moveldup_ps(inc_Val); | |
| 635 | 254 | yhp = _mm256_movehdup_ps(inc_Val); | |
| 636 | |||
| 637 | 254 | tmp1 = aVal; | |
| 638 | 254 | tmp1p = phase_Val; | |
| 639 | |||
| 640 | 254 | aVal = _mm256_shuffle_ps(aVal, aVal, 0xB1); | |
| 641 | 254 | phase_Val = _mm256_shuffle_ps(phase_Val, phase_Val, 0xB1); | |
| 642 | 254 | tmp2 = _mm256_mul_ps(aVal, yh); | |
| 643 | 254 | tmp2p = _mm256_mul_ps(phase_Val, yhp); | |
| 644 | |||
| 645 | 254 | z = _mm256_fmaddsub_ps(tmp1, yl, tmp2); | |
| 646 | 254 | phase_Val = _mm256_fmaddsub_ps(tmp1p, ylp, tmp2p); | |
| 647 | |||
| 648 | _mm256_store_ps((float*)cPtr, z); | ||
| 649 | |||
| 650 | 254 | aPtr += 4; | |
| 651 | 254 | cPtr += 4; | |
| 652 | } | ||
| 653 |
1/2✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
|
2 | if (i) { |
| 654 | 2 | tmp1 = _mm256_mul_ps(phase_Val, phase_Val); | |
| 655 | 2 | tmp2 = _mm256_hadd_ps(tmp1, tmp1); | |
| 656 | 2 | tmp1 = _mm256_shuffle_ps(tmp2, tmp2, 0xD8); | |
| 657 | 2 | tmp2 = _mm256_sqrt_ps(tmp1); | |
| 658 | 2 | phase_Val = _mm256_div_ps(phase_Val, tmp2); | |
| 659 | } | ||
| 660 | |||
| 661 | _mm256_store_ps((float*)phase_Ptr, phase_Val); | ||
| 662 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (i = 0; i < num_points % 4; ++i) { |
| 663 | 6 | *cPtr++ = *aPtr++ * phase_Ptr[0]; | |
| 664 | 6 | phase_Ptr[0] *= (phase_inc); | |
| 665 | } | ||
| 666 | |||
| 667 | 2 | (*phase) = phase_Ptr[0]; | |
| 668 | 2 | } | |
| 669 | |||
| 670 | #endif /* LV_HAVE_AVX && LV_HAVE_FMA for aligned*/ | ||
| 671 | |||
| 672 | #if LV_HAVE_AVX && LV_HAVE_FMA | ||
| 673 | #include <immintrin.h> | ||
| 674 | |||
| 675 | 2 | static inline void volk_32fc_s32fc_x2_rotator_32fc_u_avx_fma(lv_32fc_t* outVector, | |
| 676 | const lv_32fc_t* inVector, | ||
| 677 | const lv_32fc_t phase_inc, | ||
| 678 | lv_32fc_t* phase, | ||
| 679 | unsigned int num_points) | ||
| 680 | { | ||
| 681 | 2 | lv_32fc_t* cPtr = outVector; | |
| 682 | 2 | const lv_32fc_t* aPtr = inVector; | |
| 683 | 2 | lv_32fc_t incr = 1; | |
| 684 | 2 | lv_32fc_t phase_Ptr[4] = { (*phase), (*phase), (*phase), (*phase) }; | |
| 685 | |||
| 686 | 2 | unsigned int i, j = 0; | |
| 687 | |||
| 688 |
2/2✓ Branch 0 taken 8 times.
✓ Branch 1 taken 2 times.
|
10 | for (i = 0; i < 4; ++i) { |
| 689 | 8 | phase_Ptr[i] *= incr; | |
| 690 | 8 | incr *= (phase_inc); | |
| 691 | } | ||
| 692 | |||
| 693 | __m256 aVal, phase_Val, inc_Val, yl, yh, tmp1, tmp2, z, ylp, yhp, tmp1p, tmp2p; | ||
| 694 | |||
| 695 | 2 | phase_Val = _mm256_loadu_ps((float*)phase_Ptr); | |
| 696 | 2 | inc_Val = _mm256_set_ps(lv_cimag(incr), | |
| 697 | lv_creal(incr), | ||
| 698 | lv_cimag(incr), | ||
| 699 | lv_creal(incr), | ||
| 700 | lv_cimag(incr), | ||
| 701 | lv_creal(incr), | ||
| 702 | lv_cimag(incr), | ||
| 703 | lv_creal(incr)); | ||
| 704 | |||
| 705 |
2/2✓ Branch 0 taken 510 times.
✓ Branch 1 taken 2 times.
|
512 | for (i = 0; i < (unsigned int)(num_points / ROTATOR_RELOAD); i++) { |
| 706 |
2/2✓ Branch 0 taken 65280 times.
✓ Branch 1 taken 510 times.
|
65790 | for (j = 0; j < ROTATOR_RELOAD_4; ++j) { |
| 707 | |||
| 708 | 65280 | aVal = _mm256_loadu_ps((float*)aPtr); | |
| 709 | |||
| 710 | 65280 | yl = _mm256_moveldup_ps(phase_Val); | |
| 711 | 65280 | yh = _mm256_movehdup_ps(phase_Val); | |
| 712 | 65280 | ylp = _mm256_moveldup_ps(inc_Val); | |
| 713 | 65280 | yhp = _mm256_movehdup_ps(inc_Val); | |
| 714 | |||
| 715 | 65280 | tmp1 = aVal; | |
| 716 | 65280 | tmp1p = phase_Val; | |
| 717 | |||
| 718 | 65280 | aVal = _mm256_shuffle_ps(aVal, aVal, 0xB1); | |
| 719 | 65280 | phase_Val = _mm256_shuffle_ps(phase_Val, phase_Val, 0xB1); | |
| 720 | 65280 | tmp2 = _mm256_mul_ps(aVal, yh); | |
| 721 | 65280 | tmp2p = _mm256_mul_ps(phase_Val, yhp); | |
| 722 | |||
| 723 | 65280 | z = _mm256_fmaddsub_ps(tmp1, yl, tmp2); | |
| 724 | 65280 | phase_Val = _mm256_fmaddsub_ps(tmp1p, ylp, tmp2p); | |
| 725 | |||
| 726 | _mm256_storeu_ps((float*)cPtr, z); | ||
| 727 | |||
| 728 | 65280 | aPtr += 4; | |
| 729 | 65280 | cPtr += 4; | |
| 730 | } | ||
| 731 | 510 | tmp1 = _mm256_mul_ps(phase_Val, phase_Val); | |
| 732 | 510 | tmp2 = _mm256_hadd_ps(tmp1, tmp1); | |
| 733 | 510 | tmp1 = _mm256_shuffle_ps(tmp2, tmp2, 0xD8); | |
| 734 | 510 | tmp2 = _mm256_sqrt_ps(tmp1); | |
| 735 | 510 | phase_Val = _mm256_div_ps(phase_Val, tmp2); | |
| 736 | } | ||
| 737 |
2/2✓ Branch 0 taken 254 times.
✓ Branch 1 taken 2 times.
|
256 | for (i = 0; i < (num_points % ROTATOR_RELOAD) / 4; ++i) { |
| 738 | 254 | aVal = _mm256_loadu_ps((float*)aPtr); | |
| 739 | |||
| 740 | 254 | yl = _mm256_moveldup_ps(phase_Val); | |
| 741 | 254 | yh = _mm256_movehdup_ps(phase_Val); | |
| 742 | 254 | ylp = _mm256_moveldup_ps(inc_Val); | |
| 743 | 254 | yhp = _mm256_movehdup_ps(inc_Val); | |
| 744 | |||
| 745 | 254 | tmp1 = aVal; | |
| 746 | 254 | tmp1p = phase_Val; | |
| 747 | |||
| 748 | 254 | aVal = _mm256_shuffle_ps(aVal, aVal, 0xB1); | |
| 749 | 254 | phase_Val = _mm256_shuffle_ps(phase_Val, phase_Val, 0xB1); | |
| 750 | 254 | tmp2 = _mm256_mul_ps(aVal, yh); | |
| 751 | 254 | tmp2p = _mm256_mul_ps(phase_Val, yhp); | |
| 752 | |||
| 753 | 254 | z = _mm256_fmaddsub_ps(tmp1, yl, tmp2); | |
| 754 | 254 | phase_Val = _mm256_fmaddsub_ps(tmp1p, ylp, tmp2p); | |
| 755 | |||
| 756 | _mm256_storeu_ps((float*)cPtr, z); | ||
| 757 | |||
| 758 | 254 | aPtr += 4; | |
| 759 | 254 | cPtr += 4; | |
| 760 | } | ||
| 761 |
1/2✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
|
2 | if (i) { |
| 762 | 2 | tmp1 = _mm256_mul_ps(phase_Val, phase_Val); | |
| 763 | 2 | tmp2 = _mm256_hadd_ps(tmp1, tmp1); | |
| 764 | 2 | tmp1 = _mm256_shuffle_ps(tmp2, tmp2, 0xD8); | |
| 765 | 2 | tmp2 = _mm256_sqrt_ps(tmp1); | |
| 766 | 2 | phase_Val = _mm256_div_ps(phase_Val, tmp2); | |
| 767 | } | ||
| 768 | |||
| 769 | _mm256_storeu_ps((float*)phase_Ptr, phase_Val); | ||
| 770 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (i = 0; i < num_points % 4; ++i) { |
| 771 | 6 | *cPtr++ = *aPtr++ * phase_Ptr[0]; | |
| 772 | 6 | phase_Ptr[0] *= (phase_inc); | |
| 773 | } | ||
| 774 | |||
| 775 | 2 | (*phase) = phase_Ptr[0]; | |
| 776 | 2 | } | |
| 777 | |||
| 778 | #endif /* LV_HAVE_AVX && LV_HAVE_FMA*/ | ||
| 779 | |||
| 780 | #endif /* INCLUDED_volk_32fc_s32fc_rotator_32fc_a_H */ | ||
| 781 |