Vector Optimized Library of Kernels 3.1.0
Architecture-tuned implementations of math kernels
volk_32i_x2_and_32i.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: LGPL-3.0-or-later
8 */
9
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
77static 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
114static 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 unsigned int number = 0;
120 const unsigned int oneEightPoints = num_points / 8;
121
122 int32_t* cPtr = cVector;
123 const int32_t* aPtr = aVector;
124 const int32_t* bPtr = bVector;
125
126 __m256i aVal, bVal, cVal;
127 for (; number < oneEightPoints; number++) {
128
129 aVal = _mm256_load_si256((__m256i*)aPtr);
130 bVal = _mm256_load_si256((__m256i*)bPtr);
131
132 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 aPtr += 8;
138 bPtr += 8;
139 cPtr += 8;
140 }
141
142 number = oneEightPoints * 8;
143 for (; number < num_points; number++) {
144 cVector[number] = aVector[number] & bVector[number];
145 }
146}
147#endif /* LV_HAVE_AVX2 */
148
149
150#ifdef LV_HAVE_SSE
151#include <xmmintrin.h>
152
153static 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 unsigned int number = 0;
159 const unsigned int quarterPoints = num_points / 4;
160
161 float* cPtr = (float*)cVector;
162 const float* aPtr = (float*)aVector;
163 const float* bPtr = (float*)bVector;
164
165 __m128 aVal, bVal, cVal;
166 for (; number < quarterPoints; number++) {
167
168 aVal = _mm_load_ps(aPtr);
169 bVal = _mm_load_ps(bPtr);
170
171 cVal = _mm_and_ps(aVal, bVal);
172
173 _mm_store_ps(cPtr, cVal); // Store the results back into the C container
174
175 aPtr += 4;
176 bPtr += 4;
177 cPtr += 4;
178 }
179
180 number = quarterPoints * 4;
181 for (; number < num_points; number++) {
182 cVector[number] = aVector[number] & bVector[number];
183 }
184}
185#endif /* LV_HAVE_SSE */
186
187
188#ifdef LV_HAVE_NEON
189#include <arm_neon.h>
190
191static 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
223static 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 int32_t* cPtr = cVector;
229 const int32_t* aPtr = aVector;
230 const int32_t* bPtr = bVector;
231 unsigned int number = 0;
232
233 for (number = 0; number < num_points; number++) {
234 *cPtr++ = (*aPtr++) & (*bPtr++);
235 }
236}
237#endif /* LV_HAVE_GENERIC */
238
239
240#ifdef LV_HAVE_ORC
241extern 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
246static 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 volk_32i_x2_and_32i_a_orc_impl(cVector, aVector, bVector, num_points);
252}
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
268static 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
305static 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 unsigned int number = 0;
311 const unsigned int oneEightPoints = num_points / 8;
312
313 int32_t* cPtr = cVector;
314 const int32_t* aPtr = aVector;
315 const int32_t* bPtr = bVector;
316
317 __m256i aVal, bVal, cVal;
318 for (; number < oneEightPoints; number++) {
319
320 aVal = _mm256_loadu_si256((__m256i*)aPtr);
321 bVal = _mm256_loadu_si256((__m256i*)bPtr);
322
323 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 aPtr += 8;
329 bPtr += 8;
330 cPtr += 8;
331 }
332
333 number = oneEightPoints * 8;
334 for (; number < num_points; number++) {
335 cVector[number] = aVector[number] & bVector[number];
336 }
337}
338#endif /* LV_HAVE_AVX2 */
339
340
341#endif /* INCLUDED_volk_32i_x2_and_32i_u_H */