77 #ifndef INCLUDED_volk_32f_atan_32f_a_H 78 #define INCLUDED_volk_32f_atan_32f_a_H 80 #if LV_HAVE_AVX2 && LV_HAVE_FMA 81 #include <immintrin.h> 84 volk_32f_atan_32f_a_avx2_fma(
float* bVector,
const float* aVector,
unsigned int num_points)
86 float* bPtr = bVector;
87 const float* aPtr = aVector;
89 unsigned int number = 0;
90 unsigned int eighthPoints = num_points / 8;
93 __m256 aVal, pio2, x, y, z, arctangent;
94 __m256 fzeroes, fones, ftwos, ffours, condition;
96 pio2 = _mm256_set1_ps(3.14159265358979323846/2);
97 fzeroes = _mm256_setzero_ps();
98 fones = _mm256_set1_ps(1.0);
99 ftwos = _mm256_set1_ps(2.0);
100 ffours = _mm256_set1_ps(4.0);
102 for(;number < eighthPoints; number++){
103 aVal = _mm256_load_ps(aPtr);
105 condition = _mm256_cmp_ps(z, fzeroes,1);
106 z = _mm256_sub_ps(z, _mm256_and_ps(_mm256_mul_ps(z, ftwos), condition));
108 condition = _mm256_cmp_ps(z, fones,1);
109 x = _mm256_add_ps(x, _mm256_and_ps(_mm256_sub_ps(_mm256_div_ps(fones, z), z), condition));
111 for(i = 0; i < 2; i++){
112 x = _mm256_add_ps(x, _mm256_sqrt_ps(_mm256_fmadd_ps(x,x,fones)));
114 x = _mm256_div_ps(fones, x);
116 for(j =
TERMS - 1; j >=0 ; j--){
117 y = _mm256_fmadd_ps(y, _mm256_mul_ps(x, x), _mm256_set1_ps(pow(-1,j)/(2*j+1)));
120 y = _mm256_mul_ps(y, _mm256_mul_ps(x, ffours));
121 condition = _mm256_cmp_ps(z, fones,14);
123 y = _mm256_add_ps(y, _mm256_and_ps(_mm256_fnmadd_ps(y,ftwos,pio2), condition));
125 condition = _mm256_cmp_ps(aVal, fzeroes,1);
126 arctangent = _mm256_sub_ps(arctangent, _mm256_and_ps(_mm256_mul_ps(arctangent, ftwos), condition));
128 _mm256_store_ps(bPtr, arctangent);
133 number = eighthPoints * 8;
134 for(;number < num_points; number++){
135 *bPtr++ = atan(*aPtr++);
143 #include <immintrin.h> 148 float* bPtr = bVector;
149 const float* aPtr = aVector;
151 unsigned int number = 0;
152 unsigned int eighthPoints = num_points / 8;
155 __m256 aVal, pio2, x, y, z, arctangent;
156 __m256 fzeroes, fones, ftwos, ffours, condition;
158 pio2 = _mm256_set1_ps(3.14159265358979323846/2);
159 fzeroes = _mm256_setzero_ps();
160 fones = _mm256_set1_ps(1.0);
161 ftwos = _mm256_set1_ps(2.0);
162 ffours = _mm256_set1_ps(4.0);
164 for(;number < eighthPoints; number++){
165 aVal = _mm256_load_ps(aPtr);
167 condition = _mm256_cmp_ps(z, fzeroes,1);
168 z = _mm256_sub_ps(z, _mm256_and_ps(_mm256_mul_ps(z, ftwos), condition));
170 condition = _mm256_cmp_ps(z, fones,1);
171 x = _mm256_add_ps(x, _mm256_and_ps(_mm256_sub_ps(_mm256_div_ps(fones, z), z), condition));
173 for(i = 0; i < 2; i++){
174 x = _mm256_add_ps(x, _mm256_sqrt_ps(_mm256_add_ps(fones, _mm256_mul_ps(x, x))));
176 x = _mm256_div_ps(fones, x);
178 for(j =
TERMS - 1; j >=0 ; j--){
179 y = _mm256_add_ps(_mm256_mul_ps(y, _mm256_mul_ps(x, x)), _mm256_set1_ps(pow(-1,j)/(2*j+1)));
182 y = _mm256_mul_ps(y, _mm256_mul_ps(x, ffours));
183 condition = _mm256_cmp_ps(z, fones,14);
185 y = _mm256_add_ps(y, _mm256_and_ps(_mm256_sub_ps(pio2, _mm256_mul_ps(y, ftwos)), condition));
187 condition = _mm256_cmp_ps(aVal, fzeroes,1);
188 arctangent = _mm256_sub_ps(arctangent, _mm256_and_ps(_mm256_mul_ps(arctangent, ftwos), condition));
190 _mm256_store_ps(bPtr, arctangent);
195 number = eighthPoints * 8;
196 for(;number < num_points; number++){
197 *bPtr++ = atan(*aPtr++);
203 #ifdef LV_HAVE_SSE4_1 204 #include <smmintrin.h> 207 volk_32f_atan_32f_a_sse4_1(
float* bVector,
const float* aVector,
unsigned int num_points)
209 float* bPtr = bVector;
210 const float* aPtr = aVector;
212 unsigned int number = 0;
213 unsigned int quarterPoints = num_points / 4;
216 __m128 aVal, pio2, x, y, z, arctangent;
217 __m128 fzeroes, fones, ftwos, ffours, condition;
219 pio2 = _mm_set1_ps(3.14159265358979323846/2);
220 fzeroes = _mm_setzero_ps();
221 fones = _mm_set1_ps(1.0);
222 ftwos = _mm_set1_ps(2.0);
223 ffours = _mm_set1_ps(4.0);
225 for(;number < quarterPoints; number++){
226 aVal = _mm_load_ps(aPtr);
228 condition = _mm_cmplt_ps(z, fzeroes);
229 z = _mm_sub_ps(z, _mm_and_ps(_mm_mul_ps(z, ftwos), condition));
231 condition = _mm_cmplt_ps(z, fones);
232 x = _mm_add_ps(x, _mm_and_ps(_mm_sub_ps(_mm_div_ps(fones, z), z), condition));
234 for(i = 0; i < 2; i++){
235 x = _mm_add_ps(x, _mm_sqrt_ps(_mm_add_ps(fones, _mm_mul_ps(x, x))));
237 x = _mm_div_ps(fones, x);
239 for(j =
TERMS - 1; j >=0 ; j--){
240 y = _mm_add_ps(_mm_mul_ps(y, _mm_mul_ps(x, x)), _mm_set1_ps(pow(-1,j)/(2*j+1)));
243 y = _mm_mul_ps(y, _mm_mul_ps(x, ffours));
244 condition = _mm_cmpgt_ps(z, fones);
246 y = _mm_add_ps(y, _mm_and_ps(_mm_sub_ps(pio2, _mm_mul_ps(y, ftwos)), condition));
248 condition = _mm_cmplt_ps(aVal, fzeroes);
249 arctangent = _mm_sub_ps(arctangent, _mm_and_ps(_mm_mul_ps(arctangent, ftwos), condition));
251 _mm_store_ps(bPtr, arctangent);
256 number = quarterPoints * 4;
257 for(;number < num_points; number++){
258 *bPtr++ = atanf(*aPtr++);
266 #ifndef INCLUDED_volk_32f_atan_32f_u_H 267 #define INCLUDED_volk_32f_atan_32f_u_H 269 #if LV_HAVE_AVX2 && LV_HAVE_FMA 270 #include <immintrin.h> 273 volk_32f_atan_32f_u_avx2_fma(
float* bVector,
const float* aVector,
unsigned int num_points)
275 float* bPtr = bVector;
276 const float* aPtr = aVector;
278 unsigned int number = 0;
279 unsigned int eighthPoints = num_points / 8;
282 __m256 aVal, pio2, x, y, z, arctangent;
283 __m256 fzeroes, fones, ftwos, ffours, condition;
285 pio2 = _mm256_set1_ps(3.14159265358979323846/2);
286 fzeroes = _mm256_setzero_ps();
287 fones = _mm256_set1_ps(1.0);
288 ftwos = _mm256_set1_ps(2.0);
289 ffours = _mm256_set1_ps(4.0);
291 for(;number < eighthPoints; number++){
292 aVal = _mm256_loadu_ps(aPtr);
294 condition = _mm256_cmp_ps(z, fzeroes,1);
295 z = _mm256_sub_ps(z, _mm256_and_ps(_mm256_mul_ps(z, ftwos), condition));
297 condition = _mm256_cmp_ps(z, fones,1);
298 x = _mm256_add_ps(x, _mm256_and_ps(_mm256_sub_ps(_mm256_div_ps(fones, z), z), condition));
300 for(i = 0; i < 2; i++){
301 x = _mm256_add_ps(x, _mm256_sqrt_ps(_mm256_fmadd_ps(x,x,fones)));
303 x = _mm256_div_ps(fones, x);
305 for(j =
TERMS - 1; j >=0 ; j--){
306 y = _mm256_fmadd_ps(y, _mm256_mul_ps(x, x), _mm256_set1_ps(pow(-1,j)/(2*j+1)));
309 y = _mm256_mul_ps(y, _mm256_mul_ps(x, ffours));
310 condition = _mm256_cmp_ps(z, fones,14);
312 y = _mm256_add_ps(y, _mm256_and_ps(_mm256_fnmadd_ps(y,ftwos,pio2), condition));
314 condition = _mm256_cmp_ps(aVal, fzeroes,1);
315 arctangent = _mm256_sub_ps(arctangent, _mm256_and_ps(_mm256_mul_ps(arctangent, ftwos), condition));
317 _mm256_storeu_ps(bPtr, arctangent);
322 number = eighthPoints * 8;
323 for(;number < num_points; number++){
324 *bPtr++ = atan(*aPtr++);
332 #include <immintrin.h> 337 float* bPtr = bVector;
338 const float* aPtr = aVector;
340 unsigned int number = 0;
341 unsigned int eighthPoints = num_points / 8;
344 __m256 aVal, pio2, x, y, z, arctangent;
345 __m256 fzeroes, fones, ftwos, ffours, condition;
347 pio2 = _mm256_set1_ps(3.14159265358979323846/2);
348 fzeroes = _mm256_setzero_ps();
349 fones = _mm256_set1_ps(1.0);
350 ftwos = _mm256_set1_ps(2.0);
351 ffours = _mm256_set1_ps(4.0);
353 for(;number < eighthPoints; number++){
354 aVal = _mm256_loadu_ps(aPtr);
356 condition = _mm256_cmp_ps(z, fzeroes,1);
357 z = _mm256_sub_ps(z, _mm256_and_ps(_mm256_mul_ps(z, ftwos), condition));
359 condition = _mm256_cmp_ps(z, fones,1);
360 x = _mm256_add_ps(x, _mm256_and_ps(_mm256_sub_ps(_mm256_div_ps(fones, z), z), condition));
362 for(i = 0; i < 2; i++){
363 x = _mm256_add_ps(x, _mm256_sqrt_ps(_mm256_add_ps(fones, _mm256_mul_ps(x, x))));
365 x = _mm256_div_ps(fones, x);
367 for(j =
TERMS - 1; j >=0 ; j--){
368 y = _mm256_add_ps(_mm256_mul_ps(y, _mm256_mul_ps(x, x)), _mm256_set1_ps(pow(-1,j)/(2*j+1)));
371 y = _mm256_mul_ps(y, _mm256_mul_ps(x, ffours));
372 condition = _mm256_cmp_ps(z, fones,14);
374 y = _mm256_add_ps(y, _mm256_and_ps(_mm256_sub_ps(pio2, _mm256_mul_ps(y, ftwos)), condition));
376 condition = _mm256_cmp_ps(aVal, fzeroes,1);
377 arctangent = _mm256_sub_ps(arctangent, _mm256_and_ps(_mm256_mul_ps(arctangent, ftwos), condition));
379 _mm256_storeu_ps(bPtr, arctangent);
384 number = eighthPoints * 8;
385 for(;number < num_points; number++){
386 *bPtr++ = atan(*aPtr++);
392 #ifdef LV_HAVE_SSE4_1 393 #include <smmintrin.h> 396 volk_32f_atan_32f_u_sse4_1(
float* bVector,
const float* aVector,
unsigned int num_points)
398 float* bPtr = bVector;
399 const float* aPtr = aVector;
401 unsigned int number = 0;
402 unsigned int quarterPoints = num_points / 4;
405 __m128 aVal, pio2, x, y, z, arctangent;
406 __m128 fzeroes, fones, ftwos, ffours, condition;
408 pio2 = _mm_set1_ps(3.14159265358979323846/2);
409 fzeroes = _mm_setzero_ps();
410 fones = _mm_set1_ps(1.0);
411 ftwos = _mm_set1_ps(2.0);
412 ffours = _mm_set1_ps(4.0);
414 for(;number < quarterPoints; number++){
415 aVal = _mm_loadu_ps(aPtr);
417 condition = _mm_cmplt_ps(z, fzeroes);
418 z = _mm_sub_ps(z, _mm_and_ps(_mm_mul_ps(z, ftwos), condition));
420 condition = _mm_cmplt_ps(z, fones);
421 x = _mm_add_ps(x, _mm_and_ps(_mm_sub_ps(_mm_div_ps(fones, z), z), condition));
423 for(i = 0; i < 2; i++)
424 x = _mm_add_ps(x, _mm_sqrt_ps(_mm_add_ps(fones, _mm_mul_ps(x, x))));
425 x = _mm_div_ps(fones, x);
427 for(j =
TERMS - 1; j >= 0; j--)
428 y = _mm_add_ps(_mm_mul_ps(y, _mm_mul_ps(x, x)), _mm_set1_ps(pow(-1,j)/(2*j+1)));
430 y = _mm_mul_ps(y, _mm_mul_ps(x, ffours));
431 condition = _mm_cmpgt_ps(z, fones);
433 y = _mm_add_ps(y, _mm_and_ps(_mm_sub_ps(pio2, _mm_mul_ps(y, ftwos)), condition));
435 condition = _mm_cmplt_ps(aVal, fzeroes);
436 arctangent = _mm_sub_ps(arctangent, _mm_and_ps(_mm_mul_ps(arctangent, ftwos), condition));
438 _mm_storeu_ps(bPtr, arctangent);
443 number = quarterPoints * 4;
444 for(;number < num_points; number++){
445 *bPtr++ = atanf(*aPtr++);
451 #ifdef LV_HAVE_GENERIC 456 float* bPtr = bVector;
457 const float* aPtr = aVector;
458 unsigned int number = 0;
460 for(number = 0; number < num_points; number++){
461 *bPtr++ = atanf(*aPtr++);
static void volk_32f_atan_32f_u_avx(float *bVector, const float *aVector, unsigned int num_points)
Definition: volk_32f_atan_32f.h:335
static void volk_32f_atan_32f_a_avx(float *bVector, const float *aVector, unsigned int num_points)
Definition: volk_32f_atan_32f.h:146
for i
Definition: volk_config_fixed.tmpl.h:25
#define TERMS
Definition: volk_32f_atan_32f.h:75
static void volk_32f_atan_32f_generic(float *bVector, const float *aVector, unsigned int num_points)
Definition: volk_32f_atan_32f.h:454