| 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_x2_multiply_32f | ||
| 12 | * | ||
| 13 | * \b Overview | ||
| 14 | * | ||
| 15 | * Multiplies two input floating point vectors together. | ||
| 16 | * | ||
| 17 | * c[i] = a[i] * b[i] | ||
| 18 | * | ||
| 19 | * <b>Dispatcher Prototype</b> | ||
| 20 | * \code | ||
| 21 | * void volk_32f_x2_multiply_32f(float* cVector, const float* aVector, const float* | ||
| 22 | * bVector, unsigned int num_points) \endcode | ||
| 23 | * | ||
| 24 | * \b Inputs | ||
| 25 | * \li aVector: First input vector. | ||
| 26 | * \li bVector: Second input vector. | ||
| 27 | * \li num_points: The number of values in both input vectors. | ||
| 28 | * | ||
| 29 | * \b Outputs | ||
| 30 | * \li cVector: The output vector. | ||
| 31 | * | ||
| 32 | * \b Example | ||
| 33 | * Multiply elements of an increasing vector by those of a decreasing vector. | ||
| 34 | * \code | ||
| 35 | * int N = 10; | ||
| 36 | * unsigned int alignment = volk_get_alignment(); | ||
| 37 | * float* increasing = (float*)volk_malloc(sizeof(float)*N, alignment); | ||
| 38 | * float* decreasing = (float*)volk_malloc(sizeof(float)*N, alignment); | ||
| 39 | * float* out = (float*)volk_malloc(sizeof(float)*N, alignment); | ||
| 40 | * | ||
| 41 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 42 | * increasing[ii] = (float)ii; | ||
| 43 | * decreasing[ii] = 10.f - (float)ii; | ||
| 44 | * } | ||
| 45 | * | ||
| 46 | * volk_32f_x2_multiply_32f(out, increasing, decreasing, N); | ||
| 47 | * | ||
| 48 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 49 | * printf("out[%u] = %1.2f\n", ii, out[ii]); | ||
| 50 | * } | ||
| 51 | * | ||
| 52 | * volk_free(increasing); | ||
| 53 | * volk_free(decreasing); | ||
| 54 | * volk_free(out); | ||
| 55 | * \endcode | ||
| 56 | */ | ||
| 57 | |||
| 58 | #ifndef INCLUDED_volk_32f_x2_multiply_32f_u_H | ||
| 59 | #define INCLUDED_volk_32f_x2_multiply_32f_u_H | ||
| 60 | |||
| 61 | #include <inttypes.h> | ||
| 62 | #include <stdio.h> | ||
| 63 | |||
| 64 | #ifdef LV_HAVE_SSE | ||
| 65 | #include <xmmintrin.h> | ||
| 66 | |||
| 67 | 2 | static inline void volk_32f_x2_multiply_32f_u_sse(float* cVector, | |
| 68 | const float* aVector, | ||
| 69 | const float* bVector, | ||
| 70 | unsigned int num_points) | ||
| 71 | { | ||
| 72 | 2 | unsigned int number = 0; | |
| 73 | 2 | const unsigned int quarterPoints = num_points / 4; | |
| 74 | |||
| 75 | 2 | float* cPtr = cVector; | |
| 76 | 2 | const float* aPtr = aVector; | |
| 77 | 2 | const float* bPtr = bVector; | |
| 78 | |||
| 79 | __m128 aVal, bVal, cVal; | ||
| 80 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 81 | |||
| 82 | 65534 | aVal = _mm_loadu_ps(aPtr); | |
| 83 | 65534 | bVal = _mm_loadu_ps(bPtr); | |
| 84 | |||
| 85 | 65534 | cVal = _mm_mul_ps(aVal, bVal); | |
| 86 | |||
| 87 | _mm_storeu_ps(cPtr, cVal); // Store the results back into the C container | ||
| 88 | |||
| 89 | 65534 | aPtr += 4; | |
| 90 | 65534 | bPtr += 4; | |
| 91 | 65534 | cPtr += 4; | |
| 92 | } | ||
| 93 | |||
| 94 | 2 | number = quarterPoints * 4; | |
| 95 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 96 | 6 | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 97 | } | ||
| 98 | 2 | } | |
| 99 | #endif /* LV_HAVE_SSE */ | ||
| 100 | |||
| 101 | #ifdef LV_HAVE_AVX512F | ||
| 102 | #include <immintrin.h> | ||
| 103 | |||
| 104 | ✗ | static inline void volk_32f_x2_multiply_32f_u_avx512f(float* cVector, | |
| 105 | const float* aVector, | ||
| 106 | const float* bVector, | ||
| 107 | unsigned int num_points) | ||
| 108 | { | ||
| 109 | ✗ | unsigned int number = 0; | |
| 110 | ✗ | const unsigned int sixteenthPoints = num_points / 16; | |
| 111 | |||
| 112 | ✗ | float* cPtr = cVector; | |
| 113 | ✗ | const float* aPtr = aVector; | |
| 114 | ✗ | const float* bPtr = bVector; | |
| 115 | |||
| 116 | __m512 aVal, bVal, cVal; | ||
| 117 | ✗ | for (; number < sixteenthPoints; number++) { | |
| 118 | |||
| 119 | ✗ | aVal = _mm512_loadu_ps(aPtr); | |
| 120 | ✗ | bVal = _mm512_loadu_ps(bPtr); | |
| 121 | |||
| 122 | ✗ | cVal = _mm512_mul_ps(aVal, bVal); | |
| 123 | |||
| 124 | _mm512_storeu_ps(cPtr, cVal); // Store the results back into the C container | ||
| 125 | |||
| 126 | ✗ | aPtr += 16; | |
| 127 | ✗ | bPtr += 16; | |
| 128 | ✗ | cPtr += 16; | |
| 129 | } | ||
| 130 | |||
| 131 | ✗ | number = sixteenthPoints * 16; | |
| 132 | ✗ | for (; number < num_points; number++) { | |
| 133 | ✗ | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 134 | } | ||
| 135 | ✗ | } | |
| 136 | #endif /* LV_HAVE_AVX512F */ | ||
| 137 | |||
| 138 | #ifdef LV_HAVE_AVX | ||
| 139 | #include <immintrin.h> | ||
| 140 | |||
| 141 | 2 | static inline void volk_32f_x2_multiply_32f_u_avx(float* cVector, | |
| 142 | const float* aVector, | ||
| 143 | const float* bVector, | ||
| 144 | unsigned int num_points) | ||
| 145 | { | ||
| 146 | 2 | unsigned int number = 0; | |
| 147 | 2 | const unsigned int eighthPoints = num_points / 8; | |
| 148 | |||
| 149 | 2 | float* cPtr = cVector; | |
| 150 | 2 | const float* aPtr = aVector; | |
| 151 | 2 | const float* bPtr = bVector; | |
| 152 | |||
| 153 | __m256 aVal, bVal, cVal; | ||
| 154 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 155 | |||
| 156 | 32766 | aVal = _mm256_loadu_ps(aPtr); | |
| 157 | 32766 | bVal = _mm256_loadu_ps(bPtr); | |
| 158 | |||
| 159 | 32766 | cVal = _mm256_mul_ps(aVal, bVal); | |
| 160 | |||
| 161 | _mm256_storeu_ps(cPtr, cVal); // Store the results back into the C container | ||
| 162 | |||
| 163 | 32766 | aPtr += 8; | |
| 164 | 32766 | bPtr += 8; | |
| 165 | 32766 | cPtr += 8; | |
| 166 | } | ||
| 167 | |||
| 168 | 2 | number = eighthPoints * 8; | |
| 169 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 170 | 14 | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 171 | } | ||
| 172 | 2 | } | |
| 173 | #endif /* LV_HAVE_AVX */ | ||
| 174 | |||
| 175 | |||
| 176 | #ifdef LV_HAVE_GENERIC | ||
| 177 | |||
| 178 | 2 | static inline void volk_32f_x2_multiply_32f_generic(float* cVector, | |
| 179 | const float* aVector, | ||
| 180 | const float* bVector, | ||
| 181 | unsigned int num_points) | ||
| 182 | { | ||
| 183 | 2 | float* cPtr = cVector; | |
| 184 | 2 | const float* aPtr = aVector; | |
| 185 | 2 | const float* bPtr = bVector; | |
| 186 | 2 | unsigned int number = 0; | |
| 187 | |||
| 188 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
| 189 | 262142 | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 190 | } | ||
| 191 | 2 | } | |
| 192 | #endif /* LV_HAVE_GENERIC */ | ||
| 193 | |||
| 194 | |||
| 195 | #endif /* INCLUDED_volk_32f_x2_multiply_32f_u_H */ | ||
| 196 | |||
| 197 | |||
| 198 | #ifndef INCLUDED_volk_32f_x2_multiply_32f_a_H | ||
| 199 | #define INCLUDED_volk_32f_x2_multiply_32f_a_H | ||
| 200 | |||
| 201 | #include <inttypes.h> | ||
| 202 | #include <stdio.h> | ||
| 203 | |||
| 204 | #ifdef LV_HAVE_SSE | ||
| 205 | #include <xmmintrin.h> | ||
| 206 | |||
| 207 | 2 | static inline void volk_32f_x2_multiply_32f_a_sse(float* cVector, | |
| 208 | const float* aVector, | ||
| 209 | const float* bVector, | ||
| 210 | unsigned int num_points) | ||
| 211 | { | ||
| 212 | 2 | unsigned int number = 0; | |
| 213 | 2 | const unsigned int quarterPoints = num_points / 4; | |
| 214 | |||
| 215 | 2 | float* cPtr = cVector; | |
| 216 | 2 | const float* aPtr = aVector; | |
| 217 | 2 | const float* bPtr = bVector; | |
| 218 | |||
| 219 | __m128 aVal, bVal, cVal; | ||
| 220 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 221 | |||
| 222 | 65534 | aVal = _mm_load_ps(aPtr); | |
| 223 | 65534 | bVal = _mm_load_ps(bPtr); | |
| 224 | |||
| 225 | 65534 | cVal = _mm_mul_ps(aVal, bVal); | |
| 226 | |||
| 227 | _mm_store_ps(cPtr, cVal); // Store the results back into the C container | ||
| 228 | |||
| 229 | 65534 | aPtr += 4; | |
| 230 | 65534 | bPtr += 4; | |
| 231 | 65534 | cPtr += 4; | |
| 232 | } | ||
| 233 | |||
| 234 | 2 | number = quarterPoints * 4; | |
| 235 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 236 | 6 | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 237 | } | ||
| 238 | 2 | } | |
| 239 | #endif /* LV_HAVE_SSE */ | ||
| 240 | |||
| 241 | #ifdef LV_HAVE_AVX512F | ||
| 242 | #include <immintrin.h> | ||
| 243 | |||
| 244 | ✗ | static inline void volk_32f_x2_multiply_32f_a_avx512f(float* cVector, | |
| 245 | const float* aVector, | ||
| 246 | const float* bVector, | ||
| 247 | unsigned int num_points) | ||
| 248 | { | ||
| 249 | ✗ | unsigned int number = 0; | |
| 250 | ✗ | const unsigned int sixteenthPoints = num_points / 16; | |
| 251 | |||
| 252 | ✗ | float* cPtr = cVector; | |
| 253 | ✗ | const float* aPtr = aVector; | |
| 254 | ✗ | const float* bPtr = bVector; | |
| 255 | |||
| 256 | __m512 aVal, bVal, cVal; | ||
| 257 | ✗ | for (; number < sixteenthPoints; number++) { | |
| 258 | |||
| 259 | ✗ | aVal = _mm512_load_ps(aPtr); | |
| 260 | ✗ | bVal = _mm512_load_ps(bPtr); | |
| 261 | |||
| 262 | ✗ | cVal = _mm512_mul_ps(aVal, bVal); | |
| 263 | |||
| 264 | _mm512_store_ps(cPtr, cVal); // Store the results back into the C container | ||
| 265 | |||
| 266 | ✗ | aPtr += 16; | |
| 267 | ✗ | bPtr += 16; | |
| 268 | ✗ | cPtr += 16; | |
| 269 | } | ||
| 270 | |||
| 271 | ✗ | number = sixteenthPoints * 16; | |
| 272 | ✗ | for (; number < num_points; number++) { | |
| 273 | ✗ | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 274 | } | ||
| 275 | ✗ | } | |
| 276 | #endif /* LV_HAVE_AVX512F */ | ||
| 277 | |||
| 278 | |||
| 279 | #ifdef LV_HAVE_AVX | ||
| 280 | #include <immintrin.h> | ||
| 281 | |||
| 282 | 2 | static inline void volk_32f_x2_multiply_32f_a_avx(float* cVector, | |
| 283 | const float* aVector, | ||
| 284 | const float* bVector, | ||
| 285 | unsigned int num_points) | ||
| 286 | { | ||
| 287 | 2 | unsigned int number = 0; | |
| 288 | 2 | const unsigned int eighthPoints = num_points / 8; | |
| 289 | |||
| 290 | 2 | float* cPtr = cVector; | |
| 291 | 2 | const float* aPtr = aVector; | |
| 292 | 2 | const float* bPtr = bVector; | |
| 293 | |||
| 294 | __m256 aVal, bVal, cVal; | ||
| 295 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 296 | |||
| 297 | 32766 | aVal = _mm256_load_ps(aPtr); | |
| 298 | 32766 | bVal = _mm256_load_ps(bPtr); | |
| 299 | |||
| 300 | 32766 | cVal = _mm256_mul_ps(aVal, bVal); | |
| 301 | |||
| 302 | _mm256_store_ps(cPtr, cVal); // Store the results back into the C container | ||
| 303 | |||
| 304 | 32766 | aPtr += 8; | |
| 305 | 32766 | bPtr += 8; | |
| 306 | 32766 | cPtr += 8; | |
| 307 | } | ||
| 308 | |||
| 309 | 2 | number = eighthPoints * 8; | |
| 310 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 311 | 14 | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 312 | } | ||
| 313 | 2 | } | |
| 314 | #endif /* LV_HAVE_AVX */ | ||
| 315 | |||
| 316 | |||
| 317 | #ifdef LV_HAVE_NEON | ||
| 318 | #include <arm_neon.h> | ||
| 319 | |||
| 320 | static inline void volk_32f_x2_multiply_32f_neon(float* cVector, | ||
| 321 | const float* aVector, | ||
| 322 | const float* bVector, | ||
| 323 | unsigned int num_points) | ||
| 324 | { | ||
| 325 | const unsigned int quarter_points = num_points / 4; | ||
| 326 | unsigned int number; | ||
| 327 | float32x4_t avec, bvec, cvec; | ||
| 328 | for (number = 0; number < quarter_points; ++number) { | ||
| 329 | avec = vld1q_f32(aVector); | ||
| 330 | bvec = vld1q_f32(bVector); | ||
| 331 | cvec = vmulq_f32(avec, bvec); | ||
| 332 | vst1q_f32(cVector, cvec); | ||
| 333 | aVector += 4; | ||
| 334 | bVector += 4; | ||
| 335 | cVector += 4; | ||
| 336 | } | ||
| 337 | for (number = quarter_points * 4; number < num_points; ++number) { | ||
| 338 | *cVector++ = *aVector++ * *bVector++; | ||
| 339 | } | ||
| 340 | } | ||
| 341 | #endif /* LV_HAVE_NEON */ | ||
| 342 | |||
| 343 | |||
| 344 | #ifdef LV_HAVE_GENERIC | ||
| 345 | |||
| 346 | 2 | static inline void volk_32f_x2_multiply_32f_a_generic(float* cVector, | |
| 347 | const float* aVector, | ||
| 348 | const float* bVector, | ||
| 349 | unsigned int num_points) | ||
| 350 | { | ||
| 351 | 2 | float* cPtr = cVector; | |
| 352 | 2 | const float* aPtr = aVector; | |
| 353 | 2 | const float* bPtr = bVector; | |
| 354 | 2 | unsigned int number = 0; | |
| 355 | |||
| 356 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
| 357 | 262142 | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 358 | } | ||
| 359 | 2 | } | |
| 360 | #endif /* LV_HAVE_GENERIC */ | ||
| 361 | |||
| 362 | |||
| 363 | #ifdef LV_HAVE_ORC | ||
| 364 | extern void volk_32f_x2_multiply_32f_a_orc_impl(float* cVector, | ||
| 365 | const float* aVector, | ||
| 366 | const float* bVector, | ||
| 367 | unsigned int num_points); | ||
| 368 | |||
| 369 | 2 | static inline void volk_32f_x2_multiply_32f_u_orc(float* cVector, | |
| 370 | const float* aVector, | ||
| 371 | const float* bVector, | ||
| 372 | unsigned int num_points) | ||
| 373 | { | ||
| 374 | 2 | volk_32f_x2_multiply_32f_a_orc_impl(cVector, aVector, bVector, num_points); | |
| 375 | 2 | } | |
| 376 | #endif /* LV_HAVE_ORC */ | ||
| 377 | |||
| 378 | |||
| 379 | #endif /* INCLUDED_volk_32f_x2_multiply_32f_a_H */ | ||
| 380 |