57#ifndef INCLUDED_volk_32f_binary_slicer_32i_H
58#define INCLUDED_volk_32f_binary_slicer_32i_H
65 unsigned int num_points)
68 const float* aPtr = aVector;
69 unsigned int number = 0;
71 for (number = 0; number < num_points; number++) {
86 unsigned int num_points)
89 const float* aPtr = aVector;
90 unsigned int number = 0;
92 for (number = 0; number < num_points; number++) {
93 *cPtr++ = (*aPtr++ >= 0);
100#include <emmintrin.h>
103 const float* aVector,
104 unsigned int num_points)
107 const float* aPtr = aVector;
108 unsigned int number = 0;
110 unsigned int quarter_points = num_points / 4;
112 __m128i res_i, binary_i;
114 zero_val = _mm_set1_ps(0.0f);
116 for (number = 0; number < quarter_points; number++) {
117 a_val = _mm_load_ps(aPtr);
119 res_f = _mm_cmpge_ps(a_val, zero_val);
120 res_i = _mm_cvtps_epi32(res_f);
121 binary_i = _mm_srli_epi32(res_i, 31);
123 _mm_store_si128((__m128i*)cPtr, binary_i);
129 for (number = quarter_points * 4; number < num_points; number++) {
141#include <immintrin.h>
144 const float* aVector,
145 unsigned int num_points)
148 const float* aPtr = aVector;
149 unsigned int number = 0;
151 unsigned int quarter_points = num_points / 8;
152 __m256 a_val, res_f, binary_f;
154 __m256 zero_val, one_val;
155 zero_val = _mm256_set1_ps(0.0f);
156 one_val = _mm256_set1_ps(1.0f);
158 for (number = 0; number < quarter_points; number++) {
159 a_val = _mm256_load_ps(aPtr);
161 res_f = _mm256_cmp_ps(a_val, zero_val, _CMP_GE_OS);
162 binary_f = _mm256_and_ps(res_f, one_val);
163 binary_i = _mm256_cvtps_epi32(binary_f);
165 _mm256_store_si256((__m256i*)cPtr, binary_i);
171 for (number = quarter_points * 8; number < num_points; number++) {
183#include <emmintrin.h>
186 const float* aVector,
187 unsigned int num_points)
190 const float* aPtr = aVector;
191 unsigned int number = 0;
193 unsigned int quarter_points = num_points / 4;
195 __m128i res_i, binary_i;
197 zero_val = _mm_set1_ps(0.0f);
199 for (number = 0; number < quarter_points; number++) {
200 a_val = _mm_loadu_ps(aPtr);
202 res_f = _mm_cmpge_ps(a_val, zero_val);
203 res_i = _mm_cvtps_epi32(res_f);
204 binary_i = _mm_srli_epi32(res_i, 31);
206 _mm_storeu_si128((__m128i*)cPtr, binary_i);
212 for (number = quarter_points * 4; number < num_points; number++) {
224#include <immintrin.h>
227 const float* aVector,
228 unsigned int num_points)
231 const float* aPtr = aVector;
232 unsigned int number = 0;
234 unsigned int quarter_points = num_points / 8;
235 __m256 a_val, res_f, binary_f;
237 __m256 zero_val, one_val;
238 zero_val = _mm256_set1_ps(0.0f);
239 one_val = _mm256_set1_ps(1.0f);
241 for (number = 0; number < quarter_points; number++) {
242 a_val = _mm256_loadu_ps(aPtr);
244 res_f = _mm256_cmp_ps(a_val, zero_val, _CMP_GE_OS);
245 binary_f = _mm256_and_ps(res_f, one_val);
246 binary_i = _mm256_cvtps_epi32(binary_f);
248 _mm256_storeu_si256((__m256i*)cPtr, binary_i);
254 for (number = quarter_points * 8; number < num_points; number++) {
268 const float* aVector,
269 unsigned int num_points)
272 const float* aPtr = aVector;
273 unsigned int number = 0;
274 const unsigned int quarter_points = num_points / 4;
276 float32x4_t zero_val = vdupq_n_f32(0.0f);
278 for (; number < quarter_points; number++) {
279 float32x4_t a_val = vld1q_f32(aPtr);
280 uint32x4_t cmp = vcgeq_f32(a_val, zero_val);
281 uint32x4_t result = vshrq_n_u32(cmp, 31);
282 vst1q_s32(cPtr, vreinterpretq_s32_u32(result));
287 for (number = quarter_points * 4; number < num_points; number++) {
288 *cPtr++ = (*aPtr++ >= 0) ? 1 : 0;
296static inline void volk_32f_binary_slicer_32i_neonv8(
int* cVector,
297 const float* aVector,
298 unsigned int num_points)
301 const float* aPtr = aVector;
302 unsigned int number = 0;
303 const unsigned int eighth_points = num_points / 8;
305 float32x4_t zero_val = vdupq_n_f32(0.0f);
307 for (; number < eighth_points; number++) {
308 float32x4_t a_val0 = vld1q_f32(aPtr);
309 float32x4_t a_val1 = vld1q_f32(aPtr + 4);
312 uint32x4_t cmp0 = vcgeq_f32(a_val0, zero_val);
313 uint32x4_t cmp1 = vcgeq_f32(a_val1, zero_val);
314 uint32x4_t result0 = vshrq_n_u32(cmp0, 31);
315 uint32x4_t result1 = vshrq_n_u32(cmp1, 31);
317 vst1q_s32(cPtr, vreinterpretq_s32_u32(result0));
318 vst1q_s32(cPtr + 4, vreinterpretq_s32_u32(result1));
323 for (number = eighth_points * 8; number < num_points; number++) {
324 *cPtr++ = (*aPtr++ >= 0) ? 1 : 0;
330#include <riscv_vector.h>
332static inline void volk_32f_binary_slicer_32i_rvv(
int* cVector,
333 const float* aVector,
334 unsigned int num_points)
336 size_t n = num_points;
337 for (
size_t vl; n > 0; n -= vl, aVector += vl, cVector += vl) {
338 vl = __riscv_vsetvl_e32m8(n);
339 vuint32m8_t v = __riscv_vle32_v_u32m8((uint32_t*)aVector, vl);
340 v = __riscv_vsrl(__riscv_vnot(v, vl), 31, vl);
341 __riscv_vse32((uint32_t*)cVector, v, vl);