58#ifndef INCLUDED_volk_32fc_x2_multiply_conjugate_32fc_u_H
59#define INCLUDED_volk_32fc_x2_multiply_conjugate_32fc_u_H
73 unsigned int num_points)
75 unsigned int number = 0;
76 const unsigned int quarterPoints = num_points / 4;
83 for (; number < quarterPoints; number++) {
89 _mm256_storeu_ps((
float*)c, z);
96 number = quarterPoints * 4;
98 for (; number < num_points; number++) {
106#include <pmmintrin.h>
112 unsigned int num_points)
114 unsigned int number = 0;
115 const unsigned int halfPoints = num_points / 2;
122 for (; number < halfPoints; number++) {
123 x = _mm_loadu_ps((
float*)a);
124 y = _mm_loadu_ps((
float*)b);
126 _mm_storeu_ps((
float*)c, z);
133 if ((num_points % 2) != 0) {
140#ifdef LV_HAVE_GENERIC
145 unsigned int num_points)
150 unsigned int number = 0;
152 for (number = 0; number < num_points; number++) {
153 *cPtr++ = (*aPtr++) *
lv_conj(*bPtr++);
160#ifndef INCLUDED_volk_32fc_x2_multiply_conjugate_32fc_a_H
161#define INCLUDED_volk_32fc_x2_multiply_conjugate_32fc_a_H
169#include <immintrin.h>
175 unsigned int num_points)
177 unsigned int number = 0;
178 const unsigned int quarterPoints = num_points / 4;
185 for (; number < quarterPoints; number++) {
186 x = _mm256_load_ps((
float*)a);
187 y = _mm256_load_ps((
float*)b);
189 _mm256_store_ps((
float*)c, z);
196 number = quarterPoints * 4;
198 for (; number < num_points; number++) {
206#include <pmmintrin.h>
212 unsigned int num_points)
214 unsigned int number = 0;
215 const unsigned int halfPoints = num_points / 2;
222 for (; number < halfPoints; number++) {
223 x = _mm_load_ps((
float*)a);
224 y = _mm_load_ps((
float*)b);
226 _mm_store_ps((
float*)c, z);
233 if ((num_points % 2) != 0) {
246 unsigned int num_points)
250 unsigned int quarter_points = num_points / 4;
251 float32x4x2_t a_val, b_val, c_val;
252 float32x4x2_t tmp_real, tmp_imag;
253 unsigned int number = 0;
255 for (number = 0; number < quarter_points; ++number) {
256 a_val = vld2q_f32((
float*)a_ptr);
257 b_val = vld2q_f32((
float*)b_ptr);
258 b_val.val[1] = vnegq_f32(b_val.val[1]);
264 tmp_real.val[0] = vmulq_f32(a_val.val[0], b_val.val[0]);
266 tmp_real.val[1] = vmulq_f32(a_val.val[1], b_val.val[1]);
270 tmp_imag.val[0] = vmulq_f32(a_val.val[0], b_val.val[1]);
272 tmp_imag.val[1] = vmulq_f32(a_val.val[1], b_val.val[0]);
275 c_val.val[0] = vsubq_f32(tmp_real.val[0], tmp_real.val[1]);
276 c_val.val[1] = vaddq_f32(tmp_imag.val[0], tmp_imag.val[1]);
277 vst2q_f32((
float*)cVector, c_val);
284 for (number = quarter_points * 4; number < num_points; number++) {
285 *cVector++ = (*a_ptr++) * conj(*b_ptr++);
291#include <riscv_vector.h>
293static inline void volk_32fc_x2_multiply_conjugate_32fc_rvv(
lv_32fc_t* cVector,
296 unsigned int num_points)
298 size_t n = num_points;
299 for (
size_t vl; n > 0; n -= vl, aVector += vl, bVector += vl, cVector += vl) {
300 vl = __riscv_vsetvl_e32m4(n);
301 vuint64m8_t va = __riscv_vle64_v_u64m8((
const uint64_t*)aVector, vl);
302 vuint64m8_t vb = __riscv_vle64_v_u64m8((
const uint64_t*)bVector, vl);
303 vfloat32m4_t var = __riscv_vreinterpret_f32m4(__riscv_vnsrl(va, 0, vl));
304 vfloat32m4_t vbr = __riscv_vreinterpret_f32m4(__riscv_vnsrl(vb, 0, vl));
305 vfloat32m4_t vai = __riscv_vreinterpret_f32m4(__riscv_vnsrl(va, 32, vl));
306 vfloat32m4_t vbi = __riscv_vreinterpret_f32m4(__riscv_vnsrl(vb, 32, vl));
307 vfloat32m4_t vr = __riscv_vfmacc(__riscv_vfmul(var, vbr, vl), vai, vbi, vl);
308 vfloat32m4_t vi = __riscv_vfnmsac(__riscv_vfmul(vai, vbr, vl), var, vbi, vl);
309 vuint32m4_t vru = __riscv_vreinterpret_u32m4(vr);
310 vuint32m4_t viu = __riscv_vreinterpret_u32m4(vi);
312 __riscv_vwmaccu(__riscv_vwaddu_vv(vru, viu, vl), 0xFFFFFFFF, viu, vl);
313 __riscv_vse64((uint64_t*)cVector, v, vl);
319#include <riscv_vector.h>
321static inline void volk_32fc_x2_multiply_conjugate_32fc_rvvseg(
lv_32fc_t* cVector,
324 unsigned int num_points)
326 size_t n = num_points;
327 for (
size_t vl; n > 0; n -= vl, aVector += vl, bVector += vl, cVector += vl) {
328 vl = __riscv_vsetvl_e32m4(n);
329 vfloat32m4x2_t va = __riscv_vlseg2e32_v_f32m4x2((
const float*)aVector, vl);
330 vfloat32m4x2_t vb = __riscv_vlseg2e32_v_f32m4x2((
const float*)bVector, vl);
331 vfloat32m4_t var = __riscv_vget_f32m4(va, 0), vai = __riscv_vget_f32m4(va, 1);
332 vfloat32m4_t vbr = __riscv_vget_f32m4(vb, 0), vbi = __riscv_vget_f32m4(vb, 1);
333 vfloat32m4_t vr = __riscv_vfmacc(__riscv_vfmul(var, vbr, vl), vai, vbi, vl);
334 vfloat32m4_t vi = __riscv_vfnmsac(__riscv_vfmul(vai, vbr, vl), var, vbi, vl);
335 __riscv_vsseg2e32_v_f32m4x2(
336 (
float*)cVector, __riscv_vcreate_v_f32m4x2(vr, vi), vl);
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:109
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:209
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:142
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:243
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:70
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:172
static __m256 _mm256_complexconjugatemul_ps(const __m256 x, const __m256 y)
Definition volk_avx_intrinsics.h:76
#define __VOLK_PREFETCH(addr)
Definition volk_common.h:68
#define lv_conj(x)
Definition volk_complex.h:100
float complex lv_32fc_t
Definition volk_complex.h:74
static __m128 _mm_complexconjugatemul_ps(__m128 x, __m128 y)
Definition volk_sse3_intrinsics.h:31