| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* -*- c++ -*- */ | ||
| 2 | /* | ||
| 3 | * Copyright 2020 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_s32f_add_32f | ||
| 12 | * | ||
| 13 | * \b Overview | ||
| 14 | * | ||
| 15 | * Adds a floating point scalar to a floating point vector. | ||
| 16 | * | ||
| 17 | * <b>Dispatcher Prototype</b> | ||
| 18 | * \code | ||
| 19 | * void volk_32f_s32f_add_32f(float* cVector, const float* aVector, const float scalar, | ||
| 20 | * unsigned int num_points) \endcode | ||
| 21 | * | ||
| 22 | * \b Inputs | ||
| 23 | * \li aVector: The input vector of floats. | ||
| 24 | * \li scalar: the scalar value to add against \p aVector. | ||
| 25 | * \li num_points: The number of data points. | ||
| 26 | * | ||
| 27 | * \b Outputs | ||
| 28 | * \li cVector: The output vector of floats. | ||
| 29 | * | ||
| 30 | * \b Example | ||
| 31 | * \code | ||
| 32 | * int N = 10; | ||
| 33 | * unsigned int alignment = volk_get_alignment(); | ||
| 34 | * float* increasing = (float*)volk_malloc(sizeof(float)*N, alignment); | ||
| 35 | * float* out = (float*)volk_malloc(sizeof(float)*N, alignment); | ||
| 36 | * | ||
| 37 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 38 | * increasing[ii] = 2.f * ((float)ii / (float)N) - 1.f; | ||
| 39 | * } | ||
| 40 | * | ||
| 41 | * // Add addshift to each entry. | ||
| 42 | * float addshift = 5.0f; | ||
| 43 | * | ||
| 44 | * volk_32f_s32f_add_32f(out, increasing, addshift, N); | ||
| 45 | * | ||
| 46 | * for(unsigned int ii = 0; ii < N; ++ii){ | ||
| 47 | * printf("out[%u] = %f\n", ii, out[ii]); | ||
| 48 | * } | ||
| 49 | * | ||
| 50 | * volk_free(increasing); | ||
| 51 | * volk_free(out); | ||
| 52 | * \endcode | ||
| 53 | */ | ||
| 54 | |||
| 55 | #include <inttypes.h> | ||
| 56 | #include <stdio.h> | ||
| 57 | |||
| 58 | #ifndef INCLUDED_volk_32f_s32f_add_32f_u_H | ||
| 59 | #define INCLUDED_volk_32f_s32f_add_32f_u_H | ||
| 60 | |||
| 61 | #ifdef LV_HAVE_GENERIC | ||
| 62 | |||
| 63 | 10 | static inline void volk_32f_s32f_add_32f_generic(float* cVector, | |
| 64 | const float* aVector, | ||
| 65 | const float scalar, | ||
| 66 | unsigned int num_points) | ||
| 67 | { | ||
| 68 | 10 | unsigned int number = 0; | |
| 69 | 10 | const float* inputPtr = aVector; | |
| 70 | 10 | float* outputPtr = cVector; | |
| 71 |
2/2✓ Branch 0 taken 262182 times.
✓ Branch 1 taken 10 times.
|
262192 | for (number = 0; number < num_points; number++) { |
| 72 | 262182 | *outputPtr = (*inputPtr) + scalar; | |
| 73 | 262182 | inputPtr++; | |
| 74 | 262182 | outputPtr++; | |
| 75 | } | ||
| 76 | 10 | } | |
| 77 | |||
| 78 | #endif /* LV_HAVE_GENERIC */ | ||
| 79 | #ifdef LV_HAVE_SSE | ||
| 80 | #include <xmmintrin.h> | ||
| 81 | |||
| 82 | 2 | static inline void volk_32f_s32f_add_32f_u_sse(float* cVector, | |
| 83 | const float* aVector, | ||
| 84 | const float scalar, | ||
| 85 | unsigned int num_points) | ||
| 86 | { | ||
| 87 | 2 | unsigned int number = 0; | |
| 88 | 2 | const unsigned int quarterPoints = num_points / 4; | |
| 89 | |||
| 90 | 2 | float* cPtr = cVector; | |
| 91 | 2 | const float* aPtr = aVector; | |
| 92 | |||
| 93 | __m128 aVal, bVal, cVal; | ||
| 94 | 2 | bVal = _mm_set_ps1(scalar); | |
| 95 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 96 | 65534 | aVal = _mm_loadu_ps(aPtr); | |
| 97 | |||
| 98 | 65534 | cVal = _mm_add_ps(aVal, bVal); | |
| 99 | |||
| 100 | _mm_storeu_ps(cPtr, cVal); // Store the results back into the C container | ||
| 101 | |||
| 102 | 65534 | aPtr += 4; | |
| 103 | 65534 | cPtr += 4; | |
| 104 | } | ||
| 105 | |||
| 106 | 2 | number = quarterPoints * 4; | |
| 107 | 2 | volk_32f_s32f_add_32f_generic(cPtr, aPtr, scalar, num_points - number); | |
| 108 | 2 | } | |
| 109 | #endif /* LV_HAVE_SSE */ | ||
| 110 | |||
| 111 | #ifdef LV_HAVE_AVX | ||
| 112 | #include <immintrin.h> | ||
| 113 | |||
| 114 | 2 | static inline void volk_32f_s32f_add_32f_u_avx(float* cVector, | |
| 115 | const float* aVector, | ||
| 116 | const float scalar, | ||
| 117 | unsigned int num_points) | ||
| 118 | { | ||
| 119 | 2 | unsigned int number = 0; | |
| 120 | 2 | const unsigned int eighthPoints = num_points / 8; | |
| 121 | |||
| 122 | 2 | float* cPtr = cVector; | |
| 123 | 2 | const float* aPtr = aVector; | |
| 124 | |||
| 125 | __m256 aVal, bVal, cVal; | ||
| 126 | 2 | bVal = _mm256_set1_ps(scalar); | |
| 127 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 128 | |||
| 129 | 32766 | aVal = _mm256_loadu_ps(aPtr); | |
| 130 | |||
| 131 | 32766 | cVal = _mm256_add_ps(aVal, bVal); | |
| 132 | |||
| 133 | _mm256_storeu_ps(cPtr, cVal); // Store the results back into the C container | ||
| 134 | |||
| 135 | 32766 | aPtr += 8; | |
| 136 | 32766 | cPtr += 8; | |
| 137 | } | ||
| 138 | |||
| 139 | 2 | number = eighthPoints * 8; | |
| 140 | 2 | volk_32f_s32f_add_32f_generic(cPtr, aPtr, scalar, num_points - number); | |
| 141 | 2 | } | |
| 142 | #endif /* LV_HAVE_AVX */ | ||
| 143 | |||
| 144 | #ifdef LV_HAVE_NEON | ||
| 145 | #include <arm_neon.h> | ||
| 146 | |||
| 147 | static inline void volk_32f_s32f_add_32f_u_neon(float* cVector, | ||
| 148 | const float* aVector, | ||
| 149 | const float scalar, | ||
| 150 | unsigned int num_points) | ||
| 151 | { | ||
| 152 | unsigned int number = 0; | ||
| 153 | const float* inputPtr = aVector; | ||
| 154 | float* outputPtr = cVector; | ||
| 155 | const unsigned int quarterPoints = num_points / 4; | ||
| 156 | |||
| 157 | float32x4_t aVal, cVal, scalarvec; | ||
| 158 | |||
| 159 | scalarvec = vdupq_n_f32(scalar); | ||
| 160 | |||
| 161 | for (number = 0; number < quarterPoints; number++) { | ||
| 162 | aVal = vld1q_f32(inputPtr); // Load into NEON regs | ||
| 163 | cVal = vaddq_f32(aVal, scalarvec); // Do the add | ||
| 164 | vst1q_f32(outputPtr, cVal); // Store results back to output | ||
| 165 | inputPtr += 4; | ||
| 166 | outputPtr += 4; | ||
| 167 | } | ||
| 168 | |||
| 169 | number = quarterPoints * 4; | ||
| 170 | volk_32f_s32f_add_32f_generic(outputPtr, inputPtr, scalar, num_points - number); | ||
| 171 | } | ||
| 172 | #endif /* LV_HAVE_NEON */ | ||
| 173 | |||
| 174 | |||
| 175 | #endif /* INCLUDED_volk_32f_s32f_add_32f_u_H */ | ||
| 176 | |||
| 177 | |||
| 178 | #ifndef INCLUDED_volk_32f_s32f_add_32f_a_H | ||
| 179 | #define INCLUDED_volk_32f_s32f_add_32f_a_H | ||
| 180 | |||
| 181 | #ifdef LV_HAVE_SSE | ||
| 182 | #include <xmmintrin.h> | ||
| 183 | |||
| 184 | 2 | static inline void volk_32f_s32f_add_32f_a_sse(float* cVector, | |
| 185 | const float* aVector, | ||
| 186 | const float scalar, | ||
| 187 | unsigned int num_points) | ||
| 188 | { | ||
| 189 | 2 | unsigned int number = 0; | |
| 190 | 2 | const unsigned int quarterPoints = num_points / 4; | |
| 191 | |||
| 192 | 2 | float* cPtr = cVector; | |
| 193 | 2 | const float* aPtr = aVector; | |
| 194 | |||
| 195 | __m128 aVal, bVal, cVal; | ||
| 196 | 2 | bVal = _mm_set_ps1(scalar); | |
| 197 |
2/2✓ Branch 0 taken 65534 times.
✓ Branch 1 taken 2 times.
|
65536 | for (; number < quarterPoints; number++) { |
| 198 | 65534 | aVal = _mm_load_ps(aPtr); | |
| 199 | |||
| 200 | 65534 | cVal = _mm_add_ps(aVal, bVal); | |
| 201 | |||
| 202 | _mm_store_ps(cPtr, cVal); // Store the results back into the C container | ||
| 203 | |||
| 204 | 65534 | aPtr += 4; | |
| 205 | 65534 | cPtr += 4; | |
| 206 | } | ||
| 207 | |||
| 208 | 2 | number = quarterPoints * 4; | |
| 209 | 2 | volk_32f_s32f_add_32f_generic(cPtr, aPtr, scalar, num_points - number); | |
| 210 | 2 | } | |
| 211 | #endif /* LV_HAVE_SSE */ | ||
| 212 | |||
| 213 | #ifdef LV_HAVE_AVX | ||
| 214 | #include <immintrin.h> | ||
| 215 | |||
| 216 | 2 | static inline void volk_32f_s32f_add_32f_a_avx(float* cVector, | |
| 217 | const float* aVector, | ||
| 218 | const float scalar, | ||
| 219 | unsigned int num_points) | ||
| 220 | { | ||
| 221 | 2 | unsigned int number = 0; | |
| 222 | 2 | const unsigned int eighthPoints = num_points / 8; | |
| 223 | |||
| 224 | 2 | float* cPtr = cVector; | |
| 225 | 2 | const float* aPtr = aVector; | |
| 226 | |||
| 227 | __m256 aVal, bVal, cVal; | ||
| 228 | 2 | bVal = _mm256_set1_ps(scalar); | |
| 229 |
2/2✓ Branch 0 taken 32766 times.
✓ Branch 1 taken 2 times.
|
32768 | for (; number < eighthPoints; number++) { |
| 230 | 32766 | aVal = _mm256_load_ps(aPtr); | |
| 231 | |||
| 232 | 32766 | cVal = _mm256_add_ps(aVal, bVal); | |
| 233 | |||
| 234 | _mm256_store_ps(cPtr, cVal); // Store the results back into the C container | ||
| 235 | |||
| 236 | 32766 | aPtr += 8; | |
| 237 | 32766 | cPtr += 8; | |
| 238 | } | ||
| 239 | |||
| 240 | 2 | number = eighthPoints * 8; | |
| 241 | 2 | volk_32f_s32f_add_32f_generic(cPtr, aPtr, scalar, num_points - number); | |
| 242 | 2 | } | |
| 243 | #endif /* LV_HAVE_AVX */ | ||
| 244 | |||
| 245 | #ifdef LV_HAVE_ORC | ||
| 246 | |||
| 247 | extern void volk_32f_s32f_add_32f_a_orc_impl(float* dst, | ||
| 248 | const float* src, | ||
| 249 | const float scalar, | ||
| 250 | unsigned int num_points); | ||
| 251 | |||
| 252 | 2 | static inline void volk_32f_s32f_add_32f_u_orc(float* cVector, | |
| 253 | const float* aVector, | ||
| 254 | const float scalar, | ||
| 255 | unsigned int num_points) | ||
| 256 | { | ||
| 257 | 2 | volk_32f_s32f_add_32f_a_orc_impl(cVector, aVector, scalar, num_points); | |
| 258 | 2 | } | |
| 259 | #endif /* LV_HAVE_ORC */ | ||
| 260 | |||
| 261 | #endif /* INCLUDED_volk_32f_s32f_add_32f_a_H */ | ||
| 262 |