57#ifndef INCLUDED_volk_32fc_x2_multiply_32fc_u_H
58#define INCLUDED_volk_32fc_x2_multiply_32fc_u_H
65#if LV_HAVE_AVX2 && LV_HAVE_FMA
74static inline void volk_32fc_x2_multiply_32fc_u_avx2_fma(
lv_32fc_t* cVector,
77 unsigned int num_points)
79 unsigned int number = 0;
80 const unsigned int quarterPoints = num_points / 4;
86 for (; number < quarterPoints; number++) {
89 _mm256_loadu_ps((
float*)a);
91 _mm256_loadu_ps((
float*)b);
93 const __m256 yl = _mm256_moveldup_ps(y);
94 const __m256 yh = _mm256_movehdup_ps(y);
96 const __m256 tmp2x = _mm256_permute_ps(x, 0xB1);
98 const __m256 tmp2 = _mm256_mul_ps(tmp2x, yh);
100 const __m256 z = _mm256_fmaddsub_ps(
103 _mm256_storeu_ps((
float*)c, z);
110 number = quarterPoints * 4;
111 for (; number < num_points; number++) {
112 *c++ = (*a++) * (*b++);
119#include <immintrin.h>
125 unsigned int num_points)
127 unsigned int number = 0;
128 const unsigned int quarterPoints = num_points / 4;
135 for (; number < quarterPoints; number++) {
141 _mm256_storeu_ps((
float*)c, z);
148 number = quarterPoints * 4;
150 for (; number < num_points; number++) {
151 *c++ = (*a++) * (*b++);
158#include <pmmintrin.h>
164 unsigned int num_points)
166 unsigned int number = 0;
167 const unsigned int halfPoints = num_points / 2;
174 for (; number < halfPoints; number++) {
185 if ((num_points % 2) != 0) {
192#ifdef LV_HAVE_GENERIC
197 unsigned int num_points)
202 unsigned int number = 0;
204 for (number = 0; number < num_points; number++) {
205 *cPtr++ = (*aPtr++) * (*bPtr++);
212#ifndef INCLUDED_volk_32fc_x2_multiply_32fc_a_H
213#define INCLUDED_volk_32fc_x2_multiply_32fc_a_H
220#if LV_HAVE_AVX2 && LV_HAVE_FMA
221#include <immintrin.h>
229static inline void volk_32fc_x2_multiply_32fc_a_avx2_fma(
lv_32fc_t* cVector,
232 unsigned int num_points)
234 unsigned int number = 0;
235 const unsigned int quarterPoints = num_points / 4;
241 for (; number < quarterPoints; number++) {
244 _mm256_load_ps((
float*)a);
246 _mm256_load_ps((
float*)b);
248 const __m256 yl = _mm256_moveldup_ps(y);
249 const __m256 yh = _mm256_movehdup_ps(y);
251 const __m256 tmp2x = _mm256_permute_ps(x, 0xB1);
253 const __m256 tmp2 = _mm256_mul_ps(tmp2x, yh);
255 const __m256 z = _mm256_fmaddsub_ps(
258 _mm256_store_ps((
float*)c, z);
265 number = quarterPoints * 4;
266 for (; number < num_points; number++) {
267 *c++ = (*a++) * (*b++);
274#include <immintrin.h>
280 unsigned int num_points)
282 unsigned int number = 0;
283 const unsigned int quarterPoints = num_points / 4;
290 for (; number < quarterPoints; number++) {
291 x = _mm256_load_ps((
float*)a);
292 y = _mm256_load_ps((
float*)b);
294 _mm256_store_ps((
float*)c, z);
301 number = quarterPoints * 4;
303 for (; number < num_points; number++) {
304 *c++ = (*a++) * (*b++);
310#include <pmmintrin.h>
316 unsigned int num_points)
318 unsigned int number = 0;
319 const unsigned int halfPoints = num_points / 2;
326 for (; number < halfPoints; number++) {
337 if ((num_points % 2) != 0) {
344#ifdef LV_HAVE_GENERIC
349 unsigned int num_points)
354 unsigned int number = 0;
356 for (number = 0; number < num_points; number++) {
357 *cPtr++ = (*aPtr++) * (*bPtr++);
369 unsigned int num_points)
373 unsigned int quarter_points = num_points / 4;
374 float32x4x2_t a_val, b_val, c_val;
375 float32x4x2_t tmp_real, tmp_imag;
376 unsigned int number = 0;
378 for (number = 0; number < quarter_points; ++number) {
379 a_val = vld2q_f32((
float*)a_ptr);
380 b_val = vld2q_f32((
float*)b_ptr);
386 tmp_real.val[0] = vmulq_f32(a_val.val[0], b_val.val[0]);
388 tmp_real.val[1] = vmulq_f32(a_val.val[1], b_val.val[1]);
392 tmp_imag.val[0] = vmulq_f32(a_val.val[0], b_val.val[1]);
394 tmp_imag.val[1] = vmulq_f32(a_val.val[1], b_val.val[0]);
397 c_val.val[0] = vsubq_f32(tmp_real.val[0], tmp_real.val[1]);
398 c_val.val[1] = vaddq_f32(tmp_imag.val[0], tmp_imag.val[1]);
399 vst2q_f32((
float*)cVector, c_val);
406 for (number = quarter_points * 4; number < num_points; number++) {
407 *cVector++ = (*a_ptr++) * (*b_ptr++);
418 unsigned int num_points)
422 unsigned int quarter_points = num_points / 4;
423 float32x4x2_t a_val, b_val;
424 float32x4x2_t tmp_imag;
425 unsigned int number = 0;
427 for (number = 0; number < quarter_points; ++number) {
428 a_val = vld2q_f32((
float*)a_ptr);
429 b_val = vld2q_f32((
float*)b_ptr);
434 tmp_imag.val[1] = vmulq_f32(a_val.val[1], b_val.val[0]);
435 tmp_imag.val[0] = vmulq_f32(a_val.val[0], b_val.val[0]);
438 tmp_imag.val[1] = vmlaq_f32(tmp_imag.val[1], a_val.val[0], b_val.val[1]);
439 tmp_imag.val[0] = vmlsq_f32(tmp_imag.val[0], a_val.val[1], b_val.val[1]);
442 vst2q_f32((
float*)cVector, tmp_imag);
449 for (number = quarter_points * 4; number < num_points; number++) {
450 *cVector++ = (*a_ptr++) * (*b_ptr++);
458extern void volk_32fc_x2_multiply_32fc_a_neonasm(
lv_32fc_t* cVector,
461 unsigned int num_points);
467extern void volk_32fc_x2_multiply_32fc_a_orc_impl(
lv_32fc_t* cVector,
470 unsigned int num_points);
472static inline void volk_32fc_x2_multiply_32fc_u_orc(
lv_32fc_t* cVector,
475 unsigned int num_points)
477 volk_32fc_x2_multiply_32fc_a_orc_impl(cVector, aVector, bVector, num_points);