Vector Optimized Library of Kernels  2.4
Architecture-tuned implementations of math kernels
volk_32fc_x2_multiply_conjugate_32fc.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 GNU Radio
6  *
7  * GNU Radio is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License as published by
9  * the Free Software Foundation; either version 3, or (at your option)
10  * any later version.
11  *
12  * GNU Radio is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License
18  * along with GNU Radio; see the file COPYING. If not, write to
19  * the Free Software Foundation, Inc., 51 Franklin Street,
20  * Boston, MA 02110-1301, USA.
21  */
22 
71 #ifndef INCLUDED_volk_32fc_x2_multiply_conjugate_32fc_u_H
72 #define INCLUDED_volk_32fc_x2_multiply_conjugate_32fc_u_H
73 
74 #include <float.h>
75 #include <inttypes.h>
76 #include <stdio.h>
77 #include <volk/volk_complex.h>
78 
79 #ifdef LV_HAVE_AVX
80 #include <immintrin.h>
82 
84  const lv_32fc_t* aVector,
85  const lv_32fc_t* bVector,
86  unsigned int num_points)
87 {
88  unsigned int number = 0;
89  const unsigned int quarterPoints = num_points / 4;
90 
91  __m256 x, y, z;
92  lv_32fc_t* c = cVector;
93  const lv_32fc_t* a = aVector;
94  const lv_32fc_t* b = bVector;
95 
96  for (; number < quarterPoints; number++) {
97  x = _mm256_loadu_ps(
98  (float*)a); // Load the ar + ai, br + bi ... as ar,ai,br,bi ...
99  y = _mm256_loadu_ps(
100  (float*)b); // Load the cr + ci, dr + di ... as cr,ci,dr,di ...
102  _mm256_storeu_ps((float*)c, z); // Store the results back into the C container
103 
104  a += 4;
105  b += 4;
106  c += 4;
107  }
108 
109  number = quarterPoints * 4;
110 
111  for (; number < num_points; number++) {
112  *c++ = (*a++) * lv_conj(*b++);
113  }
114 }
115 #endif /* LV_HAVE_AVX */
116 
117 
118 #ifdef LV_HAVE_SSE3
119 #include <pmmintrin.h>
121 
123  const lv_32fc_t* aVector,
124  const lv_32fc_t* bVector,
125  unsigned int num_points)
126 {
127  unsigned int number = 0;
128  const unsigned int halfPoints = num_points / 2;
129 
130  __m128 x, y, z;
131  lv_32fc_t* c = cVector;
132  const lv_32fc_t* a = aVector;
133  const lv_32fc_t* b = bVector;
134 
135  for (; number < halfPoints; number++) {
136  x = _mm_loadu_ps((float*)a); // Load the ar + ai, br + bi as ar,ai,br,bi
137  y = _mm_loadu_ps((float*)b); // Load the cr + ci, dr + di as cr,ci,dr,di
138  z = _mm_complexconjugatemul_ps(x, y);
139  _mm_storeu_ps((float*)c, z); // Store the results back into the C container
140 
141  a += 2;
142  b += 2;
143  c += 2;
144  }
145 
146  if ((num_points % 2) != 0) {
147  *c = (*a) * lv_conj(*b);
148  }
149 }
150 #endif /* LV_HAVE_SSE */
151 
152 
153 #ifdef LV_HAVE_GENERIC
154 
156  const lv_32fc_t* aVector,
157  const lv_32fc_t* bVector,
158  unsigned int num_points)
159 {
160  lv_32fc_t* cPtr = cVector;
161  const lv_32fc_t* aPtr = aVector;
162  const lv_32fc_t* bPtr = bVector;
163  unsigned int number = 0;
164 
165  for (number = 0; number < num_points; number++) {
166  *cPtr++ = (*aPtr++) * lv_conj(*bPtr++);
167  }
168 }
169 #endif /* LV_HAVE_GENERIC */
170 
171 
172 #endif /* INCLUDED_volk_32fc_x2_multiply_conjugate_32fc_u_H */
173 #ifndef INCLUDED_volk_32fc_x2_multiply_conjugate_32fc_a_H
174 #define INCLUDED_volk_32fc_x2_multiply_conjugate_32fc_a_H
175 
176 #include <float.h>
177 #include <inttypes.h>
178 #include <stdio.h>
179 #include <volk/volk_complex.h>
180 
181 #ifdef LV_HAVE_AVX
182 #include <immintrin.h>
184 
186  const lv_32fc_t* aVector,
187  const lv_32fc_t* bVector,
188  unsigned int num_points)
189 {
190  unsigned int number = 0;
191  const unsigned int quarterPoints = num_points / 4;
192 
193  __m256 x, y, z;
194  lv_32fc_t* c = cVector;
195  const lv_32fc_t* a = aVector;
196  const lv_32fc_t* b = bVector;
197 
198  for (; number < quarterPoints; number++) {
199  x = _mm256_load_ps((float*)a); // Load the ar + ai, br + bi ... as ar,ai,br,bi ...
200  y = _mm256_load_ps((float*)b); // Load the cr + ci, dr + di ... as cr,ci,dr,di ...
202  _mm256_store_ps((float*)c, z); // Store the results back into the C container
203 
204  a += 4;
205  b += 4;
206  c += 4;
207  }
208 
209  number = quarterPoints * 4;
210 
211  for (; number < num_points; number++) {
212  *c++ = (*a++) * lv_conj(*b++);
213  }
214 }
215 #endif /* LV_HAVE_AVX */
216 
217 
218 #ifdef LV_HAVE_SSE3
219 #include <pmmintrin.h>
221 
223  const lv_32fc_t* aVector,
224  const lv_32fc_t* bVector,
225  unsigned int num_points)
226 {
227  unsigned int number = 0;
228  const unsigned int halfPoints = num_points / 2;
229 
230  __m128 x, y, z;
231  lv_32fc_t* c = cVector;
232  const lv_32fc_t* a = aVector;
233  const lv_32fc_t* b = bVector;
234 
235  for (; number < halfPoints; number++) {
236  x = _mm_load_ps((float*)a); // Load the ar + ai, br + bi as ar,ai,br,bi
237  y = _mm_load_ps((float*)b); // Load the cr + ci, dr + di as cr,ci,dr,di
238  z = _mm_complexconjugatemul_ps(x, y);
239  _mm_store_ps((float*)c, z); // Store the results back into the C container
240 
241  a += 2;
242  b += 2;
243  c += 2;
244  }
245 
246  if ((num_points % 2) != 0) {
247  *c = (*a) * lv_conj(*b);
248  }
249 }
250 #endif /* LV_HAVE_SSE */
251 
252 
253 #ifdef LV_HAVE_NEON
254 #include <arm_neon.h>
255 
257  const lv_32fc_t* aVector,
258  const lv_32fc_t* bVector,
259  unsigned int num_points)
260 {
261  lv_32fc_t* a_ptr = (lv_32fc_t*)aVector;
262  lv_32fc_t* b_ptr = (lv_32fc_t*)bVector;
263  unsigned int quarter_points = num_points / 4;
264  float32x4x2_t a_val, b_val, c_val;
265  float32x4x2_t tmp_real, tmp_imag;
266  unsigned int number = 0;
267 
268  for (number = 0; number < quarter_points; ++number) {
269  a_val = vld2q_f32((float*)a_ptr); // a0r|a1r|a2r|a3r || a0i|a1i|a2i|a3i
270  b_val = vld2q_f32((float*)b_ptr); // b0r|b1r|b2r|b3r || b0i|b1i|b2i|b3i
271  b_val.val[1] = vnegq_f32(b_val.val[1]);
272  __VOLK_PREFETCH(a_ptr + 4);
273  __VOLK_PREFETCH(b_ptr + 4);
274 
275  // multiply the real*real and imag*imag to get real result
276  // a0r*b0r|a1r*b1r|a2r*b2r|a3r*b3r
277  tmp_real.val[0] = vmulq_f32(a_val.val[0], b_val.val[0]);
278  // a0i*b0i|a1i*b1i|a2i*b2i|a3i*b3i
279  tmp_real.val[1] = vmulq_f32(a_val.val[1], b_val.val[1]);
280 
281  // Multiply cross terms to get the imaginary result
282  // a0r*b0i|a1r*b1i|a2r*b2i|a3r*b3i
283  tmp_imag.val[0] = vmulq_f32(a_val.val[0], b_val.val[1]);
284  // a0i*b0r|a1i*b1r|a2i*b2r|a3i*b3r
285  tmp_imag.val[1] = vmulq_f32(a_val.val[1], b_val.val[0]);
286 
287  // store the results
288  c_val.val[0] = vsubq_f32(tmp_real.val[0], tmp_real.val[1]);
289  c_val.val[1] = vaddq_f32(tmp_imag.val[0], tmp_imag.val[1]);
290  vst2q_f32((float*)cVector, c_val);
291 
292  a_ptr += 4;
293  b_ptr += 4;
294  cVector += 4;
295  }
296 
297  for (number = quarter_points * 4; number < num_points; number++) {
298  *cVector++ = (*a_ptr++) * conj(*b_ptr++);
299  }
300 }
301 #endif /* LV_HAVE_NEON */
302 
303 
304 #ifdef LV_HAVE_GENERIC
305 
306 static inline void
308  const lv_32fc_t* aVector,
309  const lv_32fc_t* bVector,
310  unsigned int num_points)
311 {
312  lv_32fc_t* cPtr = cVector;
313  const lv_32fc_t* aPtr = aVector;
314  const lv_32fc_t* bPtr = bVector;
315  unsigned int number = 0;
316 
317  for (number = 0; number < num_points; number++) {
318  *cPtr++ = (*aPtr++) * lv_conj(*bPtr++);
319  }
320 }
321 #endif /* LV_HAVE_GENERIC */
322 
323 
324 #endif /* INCLUDED_volk_32fc_x2_multiply_conjugate_32fc_a_H */
volk_32fc_x2_multiply_conjugate_32fc_neon
static void volk_32fc_x2_multiply_conjugate_32fc_neon(lv_32fc_t *cVector, const lv_32fc_t *aVector, const lv_32fc_t *bVector, unsigned int num_points)
Definition: volk_32fc_x2_multiply_conjugate_32fc.h:256
volk_sse3_intrinsics.h
volk_32fc_x2_multiply_conjugate_32fc_a_generic
static void volk_32fc_x2_multiply_conjugate_32fc_a_generic(lv_32fc_t *cVector, const lv_32fc_t *aVector, const lv_32fc_t *bVector, unsigned int num_points)
Definition: volk_32fc_x2_multiply_conjugate_32fc.h:307
volk_32fc_x2_multiply_conjugate_32fc_u_sse3
static void volk_32fc_x2_multiply_conjugate_32fc_u_sse3(lv_32fc_t *cVector, const lv_32fc_t *aVector, const lv_32fc_t *bVector, unsigned int num_points)
Definition: volk_32fc_x2_multiply_conjugate_32fc.h:122
__VOLK_PREFETCH
#define __VOLK_PREFETCH(addr)
Definition: volk_common.h:62
lv_conj
#define lv_conj(x)
Definition: volk_complex.h:96
volk_32fc_x2_multiply_conjugate_32fc_generic
static void volk_32fc_x2_multiply_conjugate_32fc_generic(lv_32fc_t *cVector, const lv_32fc_t *aVector, const lv_32fc_t *bVector, unsigned int num_points)
Definition: volk_32fc_x2_multiply_conjugate_32fc.h:155
lv_32fc_t
float complex lv_32fc_t
Definition: volk_complex.h:70
volk_complex.h
_mm256_complexconjugatemul_ps
static __m256 _mm256_complexconjugatemul_ps(const __m256 x, const __m256 y)
Definition: volk_avx_intrinsics.h:51
volk_32fc_x2_multiply_conjugate_32fc_a_avx
static void volk_32fc_x2_multiply_conjugate_32fc_a_avx(lv_32fc_t *cVector, const lv_32fc_t *aVector, const lv_32fc_t *bVector, unsigned int num_points)
Definition: volk_32fc_x2_multiply_conjugate_32fc.h:185
volk_avx_intrinsics.h
_mm_complexconjugatemul_ps
static __m128 _mm_complexconjugatemul_ps(__m128 x, __m128 y)
Definition: volk_sse3_intrinsics.h:44
volk_32fc_x2_multiply_conjugate_32fc_u_avx
static void volk_32fc_x2_multiply_conjugate_32fc_u_avx(lv_32fc_t *cVector, const lv_32fc_t *aVector, const lv_32fc_t *bVector, unsigned int num_points)
Definition: volk_32fc_x2_multiply_conjugate_32fc.h:83
volk_32fc_x2_multiply_conjugate_32fc_a_sse3
static void volk_32fc_x2_multiply_conjugate_32fc_a_sse3(lv_32fc_t *cVector, const lv_32fc_t *aVector, const lv_32fc_t *bVector, unsigned int num_points)
Definition: volk_32fc_x2_multiply_conjugate_32fc.h:222