63#ifndef INCLUDED_volk_32fc_index_min_16u_a_H
64#define INCLUDED_volk_32fc_index_min_16u_a_H
76static inline void volk_32fc_index_min_16u_a_avx2_variant_0(uint16_t* target,
80 num_points = (num_points > USHRT_MAX) ? USHRT_MAX : num_points;
82 const __m256i indices_increment = _mm256_set1_epi32(8);
88 __m256i current_indices = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0);
90 __m256 min_values = _mm256_set1_ps(FLT_MAX);
91 __m256i min_indices = _mm256_setzero_si256();
93 for (
unsigned i = 0; i < num_points / 8u; ++i) {
94 __m256 in0 = _mm256_load_ps((
float*)source);
95 __m256 in1 = _mm256_load_ps((
float*)(source + 4));
97 in0, in1, &min_values, &min_indices, ¤t_indices, indices_increment);
104 _mm256_store_ps(min_values_buffer, min_values);
105 _mm256_store_si256((__m256i*)min_indices_buffer, min_indices);
109 for (
unsigned i = 0; i < 8; i++) {
110 if (min_values_buffer[i] < min) {
111 min = min_values_buffer[i];
112 index = min_indices_buffer[i];
117 for (
unsigned i = num_points & (~7u); i < num_points; ++i) {
118 const float abs_squared =
120 if (abs_squared < min) {
133#include <immintrin.h>
136static inline void volk_32fc_index_min_16u_a_avx2_variant_1(uint16_t* target,
140 num_points = (num_points > USHRT_MAX) ? USHRT_MAX : num_points;
142 const __m256i indices_increment = _mm256_set1_epi32(8);
148 __m256i current_indices = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0);
150 __m256 min_values = _mm256_set1_ps(FLT_MAX);
151 __m256i min_indices = _mm256_setzero_si256();
153 for (
unsigned i = 0; i < num_points / 8u; ++i) {
154 __m256 in0 = _mm256_load_ps((
float*)source);
155 __m256 in1 = _mm256_load_ps((
float*)(source + 4));
157 in0, in1, &min_values, &min_indices, ¤t_indices, indices_increment);
164 _mm256_store_ps(min_values_buffer, min_values);
165 _mm256_store_si256((__m256i*)min_indices_buffer, min_indices);
169 for (
unsigned i = 0; i < 8; i++) {
170 if (min_values_buffer[i] < min) {
171 min = min_values_buffer[i];
172 index = min_indices_buffer[i];
177 for (
unsigned i = num_points & (~7u); i < num_points; ++i) {
178 const float abs_squared =
180 if (abs_squared < min) {
193#include <pmmintrin.h>
194#include <xmmintrin.h>
200 num_points = (num_points > USHRT_MAX) ? USHRT_MAX : num_points;
208 __m128i xmm8, xmm11, xmm12, xmm9, xmm10;
220 int bound = num_points >> 2;
222 for (
int i = 0;
i < bound; ++
i) {
246 if (num_points >> 1 & 1) {
273 if (num_points & 1) {
297 target[0] = holderi.
i[0];
298 sq_dist = holderf.
f[0];
299 target[0] = (holderf.
f[1] < sq_dist) ? holderi.
i[1] : target[0];
300 sq_dist = (holderf.
f[1] < sq_dist) ? holderf.
f[1] : sq_dist;
301 target[0] = (holderf.
f[2] < sq_dist) ? holderi.
i[2] : target[0];
302 sq_dist = (holderf.
f[2] < sq_dist) ? holderf.
f[2] : sq_dist;
303 target[0] = (holderf.
f[3] < sq_dist) ? holderi.
i[3] : target[0];
304 sq_dist = (holderf.
f[3] < sq_dist) ? holderf.
f[3] : sq_dist;
309#ifdef LV_HAVE_GENERIC
314 num_points = (num_points > USHRT_MAX) ? USHRT_MAX : num_points;
320 for (uint32_t
i = 0;
i < num_points; ++
i) {
336#ifndef INCLUDED_volk_32fc_index_min_16u_u_H
337#define INCLUDED_volk_32fc_index_min_16u_u_H
346#include <immintrin.h>
349static inline void volk_32fc_index_min_16u_u_avx2_variant_0(uint16_t* target,
353 num_points = (num_points > USHRT_MAX) ? USHRT_MAX : num_points;
355 const __m256i indices_increment = _mm256_set1_epi32(8);
361 __m256i current_indices = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0);
363 __m256 min_values = _mm256_set1_ps(FLT_MAX);
364 __m256i min_indices = _mm256_setzero_si256();
366 for (
unsigned i = 0;
i < num_points / 8u; ++
i) {
367 __m256 in0 = _mm256_loadu_ps((
float*)source);
368 __m256 in1 = _mm256_loadu_ps((
float*)(source + 4));
370 in0, in1, &min_values, &min_indices, ¤t_indices, indices_increment);
377 _mm256_store_ps(min_values_buffer, min_values);
378 _mm256_store_si256((__m256i*)min_indices_buffer, min_indices);
382 for (
unsigned i = 0;
i < 8;
i++) {
383 if (min_values_buffer[
i] < min) {
384 min = min_values_buffer[
i];
385 index = min_indices_buffer[
i];
390 for (
unsigned i = num_points & (~7u);
i < num_points; ++
i) {
391 const float abs_squared =
393 if (abs_squared < min) {
406#include <immintrin.h>
409static inline void volk_32fc_index_min_16u_u_avx2_variant_1(uint16_t* target,
413 num_points = (num_points > USHRT_MAX) ? USHRT_MAX : num_points;
415 const __m256i indices_increment = _mm256_set1_epi32(8);
421 __m256i current_indices = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0);
423 __m256 min_values = _mm256_set1_ps(FLT_MAX);
424 __m256i min_indices = _mm256_setzero_si256();
426 for (
unsigned i = 0;
i < num_points / 8u; ++
i) {
427 __m256 in0 = _mm256_loadu_ps((
float*)source);
428 __m256 in1 = _mm256_loadu_ps((
float*)(source + 4));
430 in0, in1, &min_values, &min_indices, ¤t_indices, indices_increment);
437 _mm256_store_ps(min_values_buffer, min_values);
438 _mm256_store_si256((__m256i*)min_indices_buffer, min_indices);
442 for (
unsigned i = 0;
i < 8;
i++) {
443 if (min_values_buffer[
i] < min) {
444 min = min_values_buffer[
i];
445 index = min_indices_buffer[
i];
450 for (
unsigned i = num_points & (~7u);
i < num_points; ++
i) {
451 const float abs_squared =
453 if (abs_squared < min) {