| 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_32i_x2_and_32i | ||
| 12 | * | ||
| 13 | * \b Overview | ||
| 14 | * | ||
| 15 | * Computes the Boolean AND operation between two input 32-bit integer vectors. | ||
| 16 | * | ||
| 17 | * <b>Dispatcher Prototype</b> | ||
| 18 | * \code | ||
| 19 | * void volk_32i_x2_and_32i(int32_t* cVector, const int32_t* aVector, const int32_t* | ||
| 20 | * bVector, unsigned int num_points) \endcode | ||
| 21 | * | ||
| 22 | * \b Inputs | ||
| 23 | * \li aVector: Input vector of samples. | ||
| 24 | * \li bVector: Input vector of samples. | ||
| 25 | * \li num_points: The number of values. | ||
| 26 | * | ||
| 27 | * \b Outputs | ||
| 28 | * \li cVector: The output vector. | ||
| 29 | * | ||
| 30 | * \b Example | ||
| 31 | * This example generates a Karnaugh map for the lower two bits of x AND y. | ||
| 32 | * \code | ||
| 33 | * int N = 1<<4; | ||
| 34 | * unsigned int alignment = volk_get_alignment(); | ||
| 35 | * | ||
| 36 | * int32_t* x = (int32_t*)volk_malloc(N*sizeof(int32_t), alignment); | ||
| 37 | * int32_t* y = (int32_t*)volk_malloc(N*sizeof(int32_t), alignment); | ||
| 38 | * int32_t* z = (int32_t*)volk_malloc(N*sizeof(int32_t), alignment); | ||
| 39 | * int32_t in_seq[] = {0,1,3,2}; | ||
| 40 | * unsigned int jj=0; | ||
| 41 | * for(unsigned int ii=0; ii<N; ++ii){ | ||
| 42 | * x[ii] = in_seq[ii%4]; | ||
| 43 | * y[ii] = in_seq[jj]; | ||
| 44 | * if(((ii+1) % 4) == 0) jj++; | ||
| 45 | * } | ||
| 46 | * | ||
| 47 | * volk_32i_x2_and_32i(z, x, y, N); | ||
| 48 | * | ||
| 49 | * printf("Karnaugh map for x AND y\n"); | ||
| 50 | * printf("y\\x|"); | ||
| 51 | * for(unsigned int ii=0; ii<4; ++ii){ | ||
| 52 | * printf(" %.2x ", in_seq[ii]); | ||
| 53 | * } | ||
| 54 | * printf("\n---|---------------\n"); | ||
| 55 | * jj = 0; | ||
| 56 | * for(unsigned int ii=0; ii<N; ++ii){ | ||
| 57 | * if(((ii+1) % 4) == 1){ | ||
| 58 | * printf("%.2x | ", in_seq[jj++]); | ||
| 59 | * } | ||
| 60 | * printf("%.2x ", z[ii]); | ||
| 61 | * if(!((ii+1) % 4)){ | ||
| 62 | * printf("\n"); | ||
| 63 | * } | ||
| 64 | * } | ||
| 65 | * \endcode | ||
| 66 | */ | ||
| 67 | |||
| 68 | #ifndef INCLUDED_volk_32i_x2_and_32i_a_H | ||
| 69 | #define INCLUDED_volk_32i_x2_and_32i_a_H | ||
| 70 | |||
| 71 | #include <inttypes.h> | ||
| 72 | #include <stdio.h> | ||
| 73 | |||
| 74 | #ifdef LV_HAVE_AVX512F | ||
| 75 | #include <immintrin.h> | ||
| 76 | |||
| 77 | ✗ | static inline void volk_32i_x2_and_32i_a_avx512f(int32_t* cVector, | |
| 78 | const int32_t* aVector, | ||
| 79 | const int32_t* bVector, | ||
| 80 | unsigned int num_points) | ||
| 81 | { | ||
| 82 | ✗ | unsigned int number = 0; | |
| 83 | ✗ | const unsigned int sixteenthPoints = num_points / 16; | |
| 84 | |||
| 85 | ✗ | int32_t* cPtr = (int32_t*)cVector; | |
| 86 | ✗ | const int32_t* aPtr = (int32_t*)aVector; | |
| 87 | ✗ | const int32_t* bPtr = (int32_t*)bVector; | |
| 88 | |||
| 89 | __m512i aVal, bVal, cVal; | ||
| 90 | ✗ | for (; number < sixteenthPoints; number++) { | |
| 91 | |||
| 92 | ✗ | aVal = _mm512_load_si512(aPtr); | |
| 93 | ✗ | bVal = _mm512_load_si512(bPtr); | |
| 94 | |||
| 95 | ✗ | cVal = _mm512_and_si512(aVal, bVal); | |
| 96 | |||
| 97 | _mm512_store_si512(cPtr, cVal); // Store the results back into the C container | ||
| 98 | |||
| 99 | ✗ | aPtr += 16; | |
| 100 | ✗ | bPtr += 16; | |
| 101 | ✗ | cPtr += 16; | |
| 102 | } | ||
| 103 | |||
| 104 | ✗ | number = sixteenthPoints * 16; | |
| 105 | ✗ | for (; number < num_points; number++) { | |
| 106 | ✗ | cVector[number] = aVector[number] & bVector[number]; | |
| 107 | } | ||
| 108 | ✗ | } | |
| 109 | #endif /* LV_HAVE_AVX512F */ | ||
| 110 | |||
| 111 | #ifdef LV_HAVE_AVX2 | ||
| 112 | #include <immintrin.h> | ||
| 113 | |||
| 114 | 2 | static inline void volk_32i_x2_and_32i_a_avx2(int32_t* cVector, | |
| 115 | const int32_t* aVector, | ||
| 116 | const int32_t* bVector, | ||
| 117 | unsigned int num_points) | ||
| 118 | { | ||
| 119 | 2 | unsigned int number = 0; | |
| 120 | 2 | const unsigned int oneEightPoints = num_points / 8; | |
| 121 | |||
| 122 | 2 | int32_t* cPtr = cVector; | |
| 123 | 2 | const int32_t* aPtr = aVector; | |
| 124 | 2 | const int32_t* bPtr = bVector; | |
| 125 | |||
| 126 | __m256i aVal, bVal, cVal; | ||
| 127 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < oneEightPoints; number++) { |
| 128 | |||
| 129 | 32766 | aVal = _mm256_load_si256((__m256i*)aPtr); | |
| 130 | 32766 | bVal = _mm256_load_si256((__m256i*)bPtr); | |
| 131 | |||
| 132 | 32766 | cVal = _mm256_and_si256(aVal, bVal); | |
| 133 | |||
| 134 | _mm256_store_si256((__m256i*)cPtr, | ||
| 135 | cVal); // Store the results back into the C container | ||
| 136 | |||
| 137 | 32766 | aPtr += 8; | |
| 138 | 32766 | bPtr += 8; | |
| 139 | 32766 | cPtr += 8; | |
| 140 | } | ||
| 141 | |||
| 142 | 2 | number = oneEightPoints * 8; | |
| 143 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 144 | 14 | cVector[number] = aVector[number] & bVector[number]; | |
| 145 | } | ||
| 146 | 2 | } | |
| 147 | #endif /* LV_HAVE_AVX2 */ | ||
| 148 | |||
| 149 | |||
| 150 | #ifdef LV_HAVE_SSE | ||
| 151 | #include <xmmintrin.h> | ||
| 152 | |||
| 153 | 2 | static inline void volk_32i_x2_and_32i_a_sse(int32_t* cVector, | |
| 154 | const int32_t* aVector, | ||
| 155 | const int32_t* bVector, | ||
| 156 | unsigned int num_points) | ||
| 157 | { | ||
| 158 | 2 | unsigned int number = 0; | |
| 159 | 2 | const unsigned int quarterPoints = num_points / 4; | |
| 160 | |||
| 161 | 2 | float* cPtr = (float*)cVector; | |
| 162 | 2 | const float* aPtr = (float*)aVector; | |
| 163 | 2 | const float* bPtr = (float*)bVector; | |
| 164 | |||
| 165 | __m128 aVal, bVal, cVal; | ||
| 166 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 167 | |||
| 168 | 65534 | aVal = _mm_load_ps(aPtr); | |
| 169 | 65534 | bVal = _mm_load_ps(bPtr); | |
| 170 | |||
| 171 | 65534 | cVal = _mm_and_ps(aVal, bVal); | |
| 172 | |||
| 173 | _mm_store_ps(cPtr, cVal); // Store the results back into the C container | ||
| 174 | |||
| 175 | 65534 | aPtr += 4; | |
| 176 | 65534 | bPtr += 4; | |
| 177 | 65534 | cPtr += 4; | |
| 178 | } | ||
| 179 | |||
| 180 | 2 | number = quarterPoints * 4; | |
| 181 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 182 | 6 | cVector[number] = aVector[number] & bVector[number]; | |
| 183 | } | ||
| 184 | 2 | } | |
| 185 | #endif /* LV_HAVE_SSE */ | ||
| 186 | |||
| 187 | |||
| 188 | #ifdef LV_HAVE_NEON | ||
| 189 | #include <arm_neon.h> | ||
| 190 | |||
| 191 | static inline void volk_32i_x2_and_32i_neon(int32_t* cVector, | ||
| 192 | const int32_t* aVector, | ||
| 193 | const int32_t* bVector, | ||
| 194 | unsigned int num_points) | ||
| 195 | { | ||
| 196 | int32_t* cPtr = cVector; | ||
| 197 | const int32_t* aPtr = aVector; | ||
| 198 | const int32_t* bPtr = bVector; | ||
| 199 | unsigned int number = 0; | ||
| 200 | unsigned int quarter_points = num_points / 4; | ||
| 201 | |||
| 202 | int32x4_t a_val, b_val, c_val; | ||
| 203 | |||
| 204 | for (number = 0; number < quarter_points; number++) { | ||
| 205 | a_val = vld1q_s32(aPtr); | ||
| 206 | b_val = vld1q_s32(bPtr); | ||
| 207 | c_val = vandq_s32(a_val, b_val); | ||
| 208 | vst1q_s32(cPtr, c_val); | ||
| 209 | aPtr += 4; | ||
| 210 | bPtr += 4; | ||
| 211 | cPtr += 4; | ||
| 212 | } | ||
| 213 | |||
| 214 | for (number = quarter_points * 4; number < num_points; number++) { | ||
| 215 | *cPtr++ = (*aPtr++) & (*bPtr++); | ||
| 216 | } | ||
| 217 | } | ||
| 218 | #endif /* LV_HAVE_NEON */ | ||
| 219 | |||
| 220 | |||
| 221 | #ifdef LV_HAVE_GENERIC | ||
| 222 | |||
| 223 | 2 | static inline void volk_32i_x2_and_32i_generic(int32_t* cVector, | |
| 224 | const int32_t* aVector, | ||
| 225 | const int32_t* bVector, | ||
| 226 | unsigned int num_points) | ||
| 227 | { | ||
| 228 | 2 | int32_t* cPtr = cVector; | |
| 229 | 2 | const int32_t* aPtr = aVector; | |
| 230 | 2 | const int32_t* bPtr = bVector; | |
| 231 | 2 | unsigned int number = 0; | |
| 232 | |||
| 233 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
| 234 | 262142 | *cPtr++ = (*aPtr++) & (*bPtr++); | |
| 235 | } | ||
| 236 | 2 | } | |
| 237 | #endif /* LV_HAVE_GENERIC */ | ||
| 238 | |||
| 239 | |||
| 240 | #ifdef LV_HAVE_ORC | ||
| 241 | extern void volk_32i_x2_and_32i_a_orc_impl(int32_t* cVector, | ||
| 242 | const int32_t* aVector, | ||
| 243 | const int32_t* bVector, | ||
| 244 | unsigned int num_points); | ||
| 245 | |||
| 246 | 2 | static inline void volk_32i_x2_and_32i_u_orc(int32_t* cVector, | |
| 247 | const int32_t* aVector, | ||
| 248 | const int32_t* bVector, | ||
| 249 | unsigned int num_points) | ||
| 250 | { | ||
| 251 | 2 | volk_32i_x2_and_32i_a_orc_impl(cVector, aVector, bVector, num_points); | |
| 252 | 2 | } | |
| 253 | #endif /* LV_HAVE_ORC */ | ||
| 254 | |||
| 255 | |||
| 256 | #endif /* INCLUDED_volk_32i_x2_and_32i_a_H */ | ||
| 257 | |||
| 258 | |||
| 259 | #ifndef INCLUDED_volk_32i_x2_and_32i_u_H | ||
| 260 | #define INCLUDED_volk_32i_x2_and_32i_u_H | ||
| 261 | |||
| 262 | #include <inttypes.h> | ||
| 263 | #include <stdio.h> | ||
| 264 | |||
| 265 | #ifdef LV_HAVE_AVX512F | ||
| 266 | #include <immintrin.h> | ||
| 267 | |||
| 268 | ✗ | static inline void volk_32i_x2_and_32i_u_avx512f(int32_t* cVector, | |
| 269 | const int32_t* aVector, | ||
| 270 | const int32_t* bVector, | ||
| 271 | unsigned int num_points) | ||
| 272 | { | ||
| 273 | ✗ | unsigned int number = 0; | |
| 274 | ✗ | const unsigned int sixteenthPoints = num_points / 16; | |
| 275 | |||
| 276 | ✗ | int32_t* cPtr = (int32_t*)cVector; | |
| 277 | ✗ | const int32_t* aPtr = (int32_t*)aVector; | |
| 278 | ✗ | const int32_t* bPtr = (int32_t*)bVector; | |
| 279 | |||
| 280 | __m512i aVal, bVal, cVal; | ||
| 281 | ✗ | for (; number < sixteenthPoints; number++) { | |
| 282 | |||
| 283 | ✗ | aVal = _mm512_loadu_si512(aPtr); | |
| 284 | ✗ | bVal = _mm512_loadu_si512(bPtr); | |
| 285 | |||
| 286 | ✗ | cVal = _mm512_and_si512(aVal, bVal); | |
| 287 | |||
| 288 | _mm512_storeu_si512(cPtr, cVal); // Store the results back into the C container | ||
| 289 | |||
| 290 | ✗ | aPtr += 16; | |
| 291 | ✗ | bPtr += 16; | |
| 292 | ✗ | cPtr += 16; | |
| 293 | } | ||
| 294 | |||
| 295 | ✗ | number = sixteenthPoints * 16; | |
| 296 | ✗ | for (; number < num_points; number++) { | |
| 297 | ✗ | cVector[number] = aVector[number] & bVector[number]; | |
| 298 | } | ||
| 299 | ✗ | } | |
| 300 | #endif /* LV_HAVE_AVX512F */ | ||
| 301 | |||
| 302 | #ifdef LV_HAVE_AVX2 | ||
| 303 | #include <immintrin.h> | ||
| 304 | |||
| 305 | 2 | static inline void volk_32i_x2_and_32i_u_avx2(int32_t* cVector, | |
| 306 | const int32_t* aVector, | ||
| 307 | const int32_t* bVector, | ||
| 308 | unsigned int num_points) | ||
| 309 | { | ||
| 310 | 2 | unsigned int number = 0; | |
| 311 | 2 | const unsigned int oneEightPoints = num_points / 8; | |
| 312 | |||
| 313 | 2 | int32_t* cPtr = cVector; | |
| 314 | 2 | const int32_t* aPtr = aVector; | |
| 315 | 2 | const int32_t* bPtr = bVector; | |
| 316 | |||
| 317 | __m256i aVal, bVal, cVal; | ||
| 318 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < oneEightPoints; number++) { |
| 319 | |||
| 320 | 32766 | aVal = _mm256_loadu_si256((__m256i*)aPtr); | |
| 321 | 32766 | bVal = _mm256_loadu_si256((__m256i*)bPtr); | |
| 322 | |||
| 323 | 32766 | cVal = _mm256_and_si256(aVal, bVal); | |
| 324 | |||
| 325 | _mm256_storeu_si256((__m256i*)cPtr, | ||
| 326 | cVal); // Store the results back into the C container | ||
| 327 | |||
| 328 | 32766 | aPtr += 8; | |
| 329 | 32766 | bPtr += 8; | |
| 330 | 32766 | cPtr += 8; | |
| 331 | } | ||
| 332 | |||
| 333 | 2 | number = oneEightPoints * 8; | |
| 334 |
2/2✓ Branch 0 taken 14 times.
✓ Branch 1 taken 2 times.
|
16 | for (; number < num_points; number++) { |
| 335 | 14 | cVector[number] = aVector[number] & bVector[number]; | |
| 336 | } | ||
| 337 | 2 | } | |
| 338 | #endif /* LV_HAVE_AVX2 */ | ||
| 339 | |||
| 340 | |||
| 341 | #endif /* INCLUDED_volk_32i_x2_and_32i_u_H */ | ||
| 342 |