| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* -*- c++ -*- */ | ||
| 2 | /* | ||
| 3 | * Copyright 2018 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_64f_x2_multiply_64f | ||
| 12 | * | ||
| 13 | * \b Overview | ||
| 14 | * | ||
| 15 | * Multiplies two input double-precision floating point vectors together. | ||
| 16 | * | ||
| 17 | * c[i] = a[i] * b[i] | ||
| 18 | * | ||
| 19 | * <b>Dispatcher Prototype</b> | ||
| 20 | * \code | ||
| 21 | * void volk_64f_x2_multiply_64f(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 | * double* increasing = (double*)volk_malloc(sizeof(double)*N, alignment); | ||
| 38 | * double* decreasing = (double*)volk_malloc(sizeof(double)*N, alignment); | ||
| 39 | * double* out = (double*)volk_malloc(sizeof(double)*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_64f_x2_multiply_64f(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_64f_x2_multiply_64f_H | ||
| 59 | #define INCLUDED_volk_64f_x2_multiply_64f_H | ||
| 60 | |||
| 61 | #include <inttypes.h> | ||
| 62 | |||
| 63 | |||
| 64 | #ifdef LV_HAVE_GENERIC | ||
| 65 | |||
| 66 | 2 | static inline void volk_64f_x2_multiply_64f_generic(double* cVector, | |
| 67 | const double* aVector, | ||
| 68 | const double* bVector, | ||
| 69 | unsigned int num_points) | ||
| 70 | { | ||
| 71 | 2 | double* cPtr = cVector; | |
| 72 | 2 | const double* aPtr = aVector; | |
| 73 | 2 | const double* bPtr = bVector; | |
| 74 | 2 | unsigned int number = 0; | |
| 75 | |||
| 76 |
2/2✓ Branch 0 taken 262142 times.
✓ Branch 1 taken 2 times.
|
262144 | for (number = 0; number < num_points; number++) { |
| 77 | 262142 | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 78 | } | ||
| 79 | 2 | } | |
| 80 | |||
| 81 | #endif /* LV_HAVE_GENERIC */ | ||
| 82 | |||
| 83 | /* | ||
| 84 | * Unaligned versions | ||
| 85 | */ | ||
| 86 | |||
| 87 | #ifdef LV_HAVE_SSE2 | ||
| 88 | |||
| 89 | #include <emmintrin.h> | ||
| 90 | |||
| 91 | 2 | static inline void volk_64f_x2_multiply_64f_u_sse2(double* cVector, | |
| 92 | const double* aVector, | ||
| 93 | const double* bVector, | ||
| 94 | unsigned int num_points) | ||
| 95 | { | ||
| 96 | 2 | unsigned int number = 0; | |
| 97 | 2 | const unsigned int half_points = num_points / 2; | |
| 98 | |||
| 99 | 2 | double* cPtr = cVector; | |
| 100 | 2 | const double* aPtr = aVector; | |
| 101 | 2 | const double* bPtr = bVector; | |
| 102 | |||
| 103 | __m128d aVal, bVal, cVal; | ||
| 104 |
2/2✓ Branch 0 taken 131070 times.
✓ Branch 1 taken 2 times.
|
131072 | for (; number < half_points; number++) { |
| 105 | 131070 | aVal = _mm_loadu_pd(aPtr); | |
| 106 | 131070 | bVal = _mm_loadu_pd(bPtr); | |
| 107 | |||
| 108 | 131070 | cVal = _mm_mul_pd(aVal, bVal); | |
| 109 | |||
| 110 | _mm_storeu_pd(cPtr, cVal); // Store the results back into the C container | ||
| 111 | |||
| 112 | 131070 | aPtr += 2; | |
| 113 | 131070 | bPtr += 2; | |
| 114 | 131070 | cPtr += 2; | |
| 115 | } | ||
| 116 | |||
| 117 | 2 | number = half_points * 2; | |
| 118 |
2/2✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
|
4 | for (; number < num_points; number++) { |
| 119 | 2 | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 120 | } | ||
| 121 | 2 | } | |
| 122 | |||
| 123 | #endif /* LV_HAVE_SSE2 */ | ||
| 124 | |||
| 125 | |||
| 126 | #ifdef LV_HAVE_AVX | ||
| 127 | |||
| 128 | #include <immintrin.h> | ||
| 129 | |||
| 130 | 2 | static inline void volk_64f_x2_multiply_64f_u_avx(double* cVector, | |
| 131 | const double* aVector, | ||
| 132 | const double* bVector, | ||
| 133 | unsigned int num_points) | ||
| 134 | { | ||
| 135 | 2 | unsigned int number = 0; | |
| 136 | 2 | const unsigned int quarter_points = num_points / 4; | |
| 137 | |||
| 138 | 2 | double* cPtr = cVector; | |
| 139 | 2 | const double* aPtr = aVector; | |
| 140 | 2 | const double* bPtr = bVector; | |
| 141 | |||
| 142 | __m256d aVal, bVal, cVal; | ||
| 143 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarter_points; number++) { |
| 144 | |||
| 145 | 65534 | aVal = _mm256_loadu_pd(aPtr); | |
| 146 | 65534 | bVal = _mm256_loadu_pd(bPtr); | |
| 147 | |||
| 148 | 65534 | cVal = _mm256_mul_pd(aVal, bVal); | |
| 149 | |||
| 150 | _mm256_storeu_pd(cPtr, cVal); // Store the results back into the C container | ||
| 151 | |||
| 152 | 65534 | aPtr += 4; | |
| 153 | 65534 | bPtr += 4; | |
| 154 | 65534 | cPtr += 4; | |
| 155 | } | ||
| 156 | |||
| 157 | 2 | number = quarter_points * 4; | |
| 158 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 159 | 6 | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 160 | } | ||
| 161 | 2 | } | |
| 162 | |||
| 163 | #endif /* LV_HAVE_AVX */ | ||
| 164 | |||
| 165 | /* | ||
| 166 | * Aligned versions | ||
| 167 | */ | ||
| 168 | |||
| 169 | #ifdef LV_HAVE_SSE2 | ||
| 170 | |||
| 171 | #include <emmintrin.h> | ||
| 172 | |||
| 173 | 2 | static inline void volk_64f_x2_multiply_64f_a_sse2(double* cVector, | |
| 174 | const double* aVector, | ||
| 175 | const double* bVector, | ||
| 176 | unsigned int num_points) | ||
| 177 | { | ||
| 178 | 2 | unsigned int number = 0; | |
| 179 | 2 | const unsigned int half_points = num_points / 2; | |
| 180 | |||
| 181 | 2 | double* cPtr = cVector; | |
| 182 | 2 | const double* aPtr = aVector; | |
| 183 | 2 | const double* bPtr = bVector; | |
| 184 | |||
| 185 | __m128d aVal, bVal, cVal; | ||
| 186 |
2/2✓ Branch 0 taken 131070 times.
✓ Branch 1 taken 2 times.
|
131072 | for (; number < half_points; number++) { |
| 187 | 131070 | aVal = _mm_load_pd(aPtr); | |
| 188 | 131070 | bVal = _mm_load_pd(bPtr); | |
| 189 | |||
| 190 | 131070 | cVal = _mm_mul_pd(aVal, bVal); | |
| 191 | |||
| 192 | _mm_store_pd(cPtr, cVal); // Store the results back into the C container | ||
| 193 | |||
| 194 | 131070 | aPtr += 2; | |
| 195 | 131070 | bPtr += 2; | |
| 196 | 131070 | cPtr += 2; | |
| 197 | } | ||
| 198 | |||
| 199 | 2 | number = half_points * 2; | |
| 200 |
2/2✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
|
4 | for (; number < num_points; number++) { |
| 201 | 2 | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 202 | } | ||
| 203 | 2 | } | |
| 204 | |||
| 205 | #endif /* LV_HAVE_SSE2 */ | ||
| 206 | |||
| 207 | |||
| 208 | #ifdef LV_HAVE_AVX | ||
| 209 | |||
| 210 | #include <immintrin.h> | ||
| 211 | |||
| 212 | 2 | static inline void volk_64f_x2_multiply_64f_a_avx(double* cVector, | |
| 213 | const double* aVector, | ||
| 214 | const double* bVector, | ||
| 215 | unsigned int num_points) | ||
| 216 | { | ||
| 217 | 2 | unsigned int number = 0; | |
| 218 | 2 | const unsigned int quarter_points = num_points / 4; | |
| 219 | |||
| 220 | 2 | double* cPtr = cVector; | |
| 221 | 2 | const double* aPtr = aVector; | |
| 222 | 2 | const double* bPtr = bVector; | |
| 223 | |||
| 224 | __m256d aVal, bVal, cVal; | ||
| 225 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarter_points; number++) { |
| 226 | |||
| 227 | 65534 | aVal = _mm256_load_pd(aPtr); | |
| 228 | 65534 | bVal = _mm256_load_pd(bPtr); | |
| 229 | |||
| 230 | 65534 | cVal = _mm256_mul_pd(aVal, bVal); | |
| 231 | |||
| 232 | _mm256_store_pd(cPtr, cVal); // Store the results back into the C container | ||
| 233 | |||
| 234 | 65534 | aPtr += 4; | |
| 235 | 65534 | bPtr += 4; | |
| 236 | 65534 | cPtr += 4; | |
| 237 | } | ||
| 238 | |||
| 239 | 2 | number = quarter_points * 4; | |
| 240 |
2/2✓ Branch 0 taken 6 times.
✓ Branch 1 taken 2 times.
|
8 | for (; number < num_points; number++) { |
| 241 | 6 | *cPtr++ = (*aPtr++) * (*bPtr++); | |
| 242 | } | ||
| 243 | 2 | } | |
| 244 | |||
| 245 | #endif /* LV_HAVE_AVX */ | ||
| 246 | |||
| 247 | #endif /* INCLUDED_volk_64f_x2_multiply_64f_u_H */ | ||
| 248 |