Vector Optimized Library of Kernels 2.5.2
Architecture-tuned implementations of math kernels
volk_32f_s32f_multiply_32f.h
Go to the documentation of this file.
1/* -*- c++ -*- */
2/*
3 * Copyright 2012, 2014 Free Software Foundation, Inc.
4 *
5 * This file is part of VOLK
6 *
7 * SPDX-License-Identifier: GPL-3.0-or-later
8 */
9
56#ifndef INCLUDED_volk_32f_s32f_multiply_32f_u_H
57#define INCLUDED_volk_32f_s32f_multiply_32f_u_H
58
59#include <inttypes.h>
60#include <stdio.h>
61
62#ifdef LV_HAVE_SSE
63#include <xmmintrin.h>
64
65static inline void volk_32f_s32f_multiply_32f_u_sse(float* cVector,
66 const float* aVector,
67 const float scalar,
68 unsigned int num_points)
69{
70 unsigned int number = 0;
71 const unsigned int quarterPoints = num_points / 4;
72
73 float* cPtr = cVector;
74 const float* aPtr = aVector;
75
76 __m128 aVal, bVal, cVal;
77 bVal = _mm_set_ps1(scalar);
78 for (; number < quarterPoints; number++) {
79 aVal = _mm_loadu_ps(aPtr);
80
81 cVal = _mm_mul_ps(aVal, bVal);
82
83 _mm_storeu_ps(cPtr, cVal); // Store the results back into the C container
84
85 aPtr += 4;
86 cPtr += 4;
87 }
88
89 number = quarterPoints * 4;
90 for (; number < num_points; number++) {
91 *cPtr++ = (*aPtr++) * scalar;
92 }
93}
94#endif /* LV_HAVE_SSE */
95
96#ifdef LV_HAVE_AVX
97#include <immintrin.h>
98
99static inline void volk_32f_s32f_multiply_32f_u_avx(float* cVector,
100 const float* aVector,
101 const float scalar,
102 unsigned int num_points)
103{
104 unsigned int number = 0;
105 const unsigned int eighthPoints = num_points / 8;
106
107 float* cPtr = cVector;
108 const float* aPtr = aVector;
109
110 __m256 aVal, bVal, cVal;
111 bVal = _mm256_set1_ps(scalar);
112 for (; number < eighthPoints; number++) {
113
114 aVal = _mm256_loadu_ps(aPtr);
115
116 cVal = _mm256_mul_ps(aVal, bVal);
117
118 _mm256_storeu_ps(cPtr, cVal); // Store the results back into the C container
119
120 aPtr += 8;
121 cPtr += 8;
122 }
123
124 number = eighthPoints * 8;
125 for (; number < num_points; number++) {
126 *cPtr++ = (*aPtr++) * scalar;
127 }
128}
129#endif /* LV_HAVE_AVX */
130
131#ifdef LV_HAVE_GENERIC
132
133static inline void volk_32f_s32f_multiply_32f_generic(float* cVector,
134 const float* aVector,
135 const float scalar,
136 unsigned int num_points)
137{
138 unsigned int number = 0;
139 const float* inputPtr = aVector;
140 float* outputPtr = cVector;
141 for (number = 0; number < num_points; number++) {
142 *outputPtr = (*inputPtr) * scalar;
143 inputPtr++;
144 outputPtr++;
145 }
146}
147#endif /* LV_HAVE_GENERIC */
148
149#endif /* INCLUDED_volk_32f_s32f_multiply_32f_u_H */
150
151
152#ifndef INCLUDED_volk_32f_s32f_multiply_32f_a_H
153#define INCLUDED_volk_32f_s32f_multiply_32f_a_H
154
155#include <inttypes.h>
156#include <stdio.h>
157
158#ifdef LV_HAVE_SSE
159#include <xmmintrin.h>
160
161static inline void volk_32f_s32f_multiply_32f_a_sse(float* cVector,
162 const float* aVector,
163 const float scalar,
164 unsigned int num_points)
165{
166 unsigned int number = 0;
167 const unsigned int quarterPoints = num_points / 4;
168
169 float* cPtr = cVector;
170 const float* aPtr = aVector;
171
172 __m128 aVal, bVal, cVal;
173 bVal = _mm_set_ps1(scalar);
174 for (; number < quarterPoints; number++) {
175 aVal = _mm_load_ps(aPtr);
176
177 cVal = _mm_mul_ps(aVal, bVal);
178
179 _mm_store_ps(cPtr, cVal); // Store the results back into the C container
180
181 aPtr += 4;
182 cPtr += 4;
183 }
184
185 number = quarterPoints * 4;
186 for (; number < num_points; number++) {
187 *cPtr++ = (*aPtr++) * scalar;
188 }
189}
190#endif /* LV_HAVE_SSE */
191
192#ifdef LV_HAVE_AVX
193#include <immintrin.h>
194
195static inline void volk_32f_s32f_multiply_32f_a_avx(float* cVector,
196 const float* aVector,
197 const float scalar,
198 unsigned int num_points)
199{
200 unsigned int number = 0;
201 const unsigned int eighthPoints = num_points / 8;
202
203 float* cPtr = cVector;
204 const float* aPtr = aVector;
205
206 __m256 aVal, bVal, cVal;
207 bVal = _mm256_set1_ps(scalar);
208 for (; number < eighthPoints; number++) {
209 aVal = _mm256_load_ps(aPtr);
210
211 cVal = _mm256_mul_ps(aVal, bVal);
212
213 _mm256_store_ps(cPtr, cVal); // Store the results back into the C container
214
215 aPtr += 8;
216 cPtr += 8;
217 }
218
219 number = eighthPoints * 8;
220 for (; number < num_points; number++) {
221 *cPtr++ = (*aPtr++) * scalar;
222 }
223}
224#endif /* LV_HAVE_AVX */
225
226#ifdef LV_HAVE_NEON
227#include <arm_neon.h>
228
229static inline void volk_32f_s32f_multiply_32f_u_neon(float* cVector,
230 const float* aVector,
231 const float scalar,
232 unsigned int num_points)
233{
234 unsigned int number = 0;
235 const float* inputPtr = aVector;
236 float* outputPtr = cVector;
237 const unsigned int quarterPoints = num_points / 4;
238
239 float32x4_t aVal, cVal;
240
241 for (number = 0; number < quarterPoints; number++) {
242 aVal = vld1q_f32(inputPtr); // Load into NEON regs
243 cVal = vmulq_n_f32(aVal, scalar); // Do the multiply
244 vst1q_f32(outputPtr, cVal); // Store results back to output
245 inputPtr += 4;
246 outputPtr += 4;
247 }
248 for (number = quarterPoints * 4; number < num_points; number++) {
249 *outputPtr++ = (*inputPtr++) * scalar;
250 }
251}
252#endif /* LV_HAVE_NEON */
253
254
255#ifdef LV_HAVE_GENERIC
256
257static inline void volk_32f_s32f_multiply_32f_a_generic(float* cVector,
258 const float* aVector,
259 const float scalar,
260 unsigned int num_points)
261{
262 unsigned int number = 0;
263 const float* inputPtr = aVector;
264 float* outputPtr = cVector;
265 for (number = 0; number < num_points; number++) {
266 *outputPtr = (*inputPtr) * scalar;
267 inputPtr++;
268 outputPtr++;
269 }
270}
271#endif /* LV_HAVE_GENERIC */
272
273
274#ifdef LV_HAVE_ORC
275
276extern void volk_32f_s32f_multiply_32f_a_orc_impl(float* dst,
277 const float* src,
278 const float scalar,
279 unsigned int num_points);
280
281static inline void volk_32f_s32f_multiply_32f_u_orc(float* cVector,
282 const float* aVector,
283 const float scalar,
284 unsigned int num_points)
285{
286 volk_32f_s32f_multiply_32f_a_orc_impl(cVector, aVector, scalar, num_points);
287}
288
289#endif /* LV_HAVE_GENERIC */
290
291#endif /* INCLUDED_volk_32f_s32f_multiply_32f_a_H */