60#ifndef INCLUDED_volk_32fc_s32f_magnitude_16i_a_H
61#define INCLUDED_volk_32fc_s32f_magnitude_16i_a_H
73 unsigned int num_points)
75 const float* complexVectorPtr = (
float*)complexVector;
76 int16_t* magnitudeVectorPtr = magnitudeVector;
77 unsigned int number = 0;
78 for (number = 0; number < num_points; number++) {
79 float real = *complexVectorPtr++;
80 float imag = *complexVectorPtr++;
81 *magnitudeVectorPtr++ =
82 (int16_t)
rintf(scalar * sqrtf((real * real) + (imag * imag)));
90static inline void volk_32fc_s32f_magnitude_16i_a_avx2(int16_t* magnitudeVector,
93 unsigned int num_points)
95 unsigned int number = 0;
96 const unsigned int eighthPoints = num_points / 8;
98 const float* complexVectorPtr = (
const float*)complexVector;
99 int16_t* magnitudeVectorPtr = magnitudeVector;
101 __m256 vScalar = _mm256_set1_ps(scalar);
102 __m256i idx = _mm256_set_epi32(0, 0, 0, 0, 5, 1, 4, 0);
103 __m256 cplxValue1, cplxValue2, result;
107 for (; number < eighthPoints; number++) {
108 cplxValue1 = _mm256_load_ps(complexVectorPtr);
109 complexVectorPtr += 8;
111 cplxValue2 = _mm256_load_ps(complexVectorPtr);
112 complexVectorPtr += 8;
114 cplxValue1 = _mm256_mul_ps(cplxValue1, cplxValue1);
115 cplxValue2 = _mm256_mul_ps(cplxValue2, cplxValue2);
117 result = _mm256_hadd_ps(cplxValue1, cplxValue2);
119 result = _mm256_sqrt_ps(result);
121 result = _mm256_mul_ps(result, vScalar);
123 resultInt = _mm256_cvtps_epi32(result);
124 resultInt = _mm256_packs_epi32(resultInt, resultInt);
125 resultInt = _mm256_permutevar8x32_epi32(
127 resultShort = _mm256_extracti128_si256(resultInt, 0);
128 _mm_store_si128((__m128i*)magnitudeVectorPtr, resultShort);
129 magnitudeVectorPtr += 8;
132 number = eighthPoints * 8;
134 magnitudeVector + number, complexVector + number, scalar, num_points - number);
139#include <pmmintrin.h>
144 unsigned int num_points)
146 unsigned int number = 0;
147 const unsigned int quarterPoints = num_points / 4;
149 const float* complexVectorPtr = (
const float*)complexVector;
150 int16_t* magnitudeVectorPtr = magnitudeVector;
152 __m128 vScalar = _mm_set_ps1(scalar);
154 __m128 cplxValue1, cplxValue2, result;
158 for (; number < quarterPoints; number++) {
159 cplxValue1 = _mm_load_ps(complexVectorPtr);
160 complexVectorPtr += 4;
162 cplxValue2 = _mm_load_ps(complexVectorPtr);
163 complexVectorPtr += 4;
165 cplxValue1 = _mm_mul_ps(cplxValue1, cplxValue1);
166 cplxValue2 = _mm_mul_ps(cplxValue2, cplxValue2);
168 result = _mm_hadd_ps(cplxValue1, cplxValue2);
170 result = _mm_sqrt_ps(result);
172 result = _mm_mul_ps(result, vScalar);
174 _mm_store_ps(floatBuffer, result);
175 *magnitudeVectorPtr++ = (int16_t)
rintf(floatBuffer[0]);
176 *magnitudeVectorPtr++ = (int16_t)
rintf(floatBuffer[1]);
177 *magnitudeVectorPtr++ = (int16_t)
rintf(floatBuffer[2]);
178 *magnitudeVectorPtr++ = (int16_t)
rintf(floatBuffer[3]);
181 number = quarterPoints * 4;
183 magnitudeVector + number, complexVector + number, scalar, num_points - number);
189#include <xmmintrin.h>
194 unsigned int num_points)
196 unsigned int number = 0;
197 const unsigned int quarterPoints = num_points / 4;
199 const float* complexVectorPtr = (
const float*)complexVector;
200 int16_t* magnitudeVectorPtr = magnitudeVector;
202 __m128 vScalar = _mm_set_ps1(scalar);
204 __m128 cplxValue1, cplxValue2, result;
205 __m128 iValue, qValue;
209 for (; number < quarterPoints; number++) {
210 cplxValue1 = _mm_load_ps(complexVectorPtr);
211 complexVectorPtr += 4;
213 cplxValue2 = _mm_load_ps(complexVectorPtr);
214 complexVectorPtr += 4;
217 iValue = _mm_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(2, 0, 2, 0));
219 qValue = _mm_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(3, 1, 3, 1));
221 __m128 iValue2 = _mm_mul_ps(iValue, iValue);
222 __m128 qValue2 = _mm_mul_ps(qValue, qValue);
224 result = _mm_add_ps(iValue2, qValue2);
226 result = _mm_sqrt_ps(result);
228 result = _mm_mul_ps(result, vScalar);
230 _mm_store_ps(floatBuffer, result);
231 *magnitudeVectorPtr++ = (int16_t)
rintf(floatBuffer[0]);
232 *magnitudeVectorPtr++ = (int16_t)
rintf(floatBuffer[1]);
233 *magnitudeVectorPtr++ = (int16_t)
rintf(floatBuffer[2]);
234 *magnitudeVectorPtr++ = (int16_t)
rintf(floatBuffer[3]);
237 number = quarterPoints * 4;
239 magnitudeVector + number, complexVector + number, scalar, num_points - number);
246#ifndef INCLUDED_volk_32fc_s32f_magnitude_16i_u_H
247#define INCLUDED_volk_32fc_s32f_magnitude_16i_u_H
255#include <immintrin.h>
257static inline void volk_32fc_s32f_magnitude_16i_u_avx2(int16_t* magnitudeVector,
260 unsigned int num_points)
262 unsigned int number = 0;
263 const unsigned int eighthPoints = num_points / 8;
265 const float* complexVectorPtr = (
const float*)complexVector;
266 int16_t* magnitudeVectorPtr = magnitudeVector;
268 __m256 vScalar = _mm256_set1_ps(scalar);
269 __m256i idx = _mm256_set_epi32(0, 0, 0, 0, 5, 1, 4, 0);
270 __m256 cplxValue1, cplxValue2, result;
274 for (; number < eighthPoints; number++) {
275 cplxValue1 = _mm256_loadu_ps(complexVectorPtr);
276 complexVectorPtr += 8;
278 cplxValue2 = _mm256_loadu_ps(complexVectorPtr);
279 complexVectorPtr += 8;
281 cplxValue1 = _mm256_mul_ps(cplxValue1, cplxValue1);
282 cplxValue2 = _mm256_mul_ps(cplxValue2, cplxValue2);
284 result = _mm256_hadd_ps(cplxValue1, cplxValue2);
286 result = _mm256_sqrt_ps(result);
288 result = _mm256_mul_ps(result, vScalar);
290 resultInt = _mm256_cvtps_epi32(result);
291 resultInt = _mm256_packs_epi32(resultInt, resultInt);
292 resultInt = _mm256_permutevar8x32_epi32(
294 resultShort = _mm256_extracti128_si256(resultInt, 0);
295 _mm_storeu_si128((__m128i*)magnitudeVectorPtr, resultShort);
296 magnitudeVectorPtr += 8;
299 number = eighthPoints * 8;
301 magnitudeVector + number, complexVector + number, scalar, num_points - number);
306#include <riscv_vector.h>
308static inline void volk_32fc_s32f_magnitude_16i_rvv(int16_t* magnitudeVector,
311 unsigned int num_points)
313 size_t n = num_points;
314 for (
size_t vl; n > 0; n -= vl, complexVector += vl, magnitudeVector += vl) {
315 vl = __riscv_vsetvl_e32m4(n);
316 vuint64m8_t vc = __riscv_vle64_v_u64m8((
const uint64_t*)complexVector, vl);
317 vfloat32m4_t vr = __riscv_vreinterpret_f32m4(__riscv_vnsrl(vc, 0, vl));
318 vfloat32m4_t vi = __riscv_vreinterpret_f32m4(__riscv_vnsrl(vc, 32, vl));
319 vfloat32m4_t v = __riscv_vfmacc(__riscv_vfmul(vi, vi, vl), vr, vr, vl);
320 v = __riscv_vfmul(__riscv_vfsqrt(v, vl), scalar, vl);
321 __riscv_vse16(magnitudeVector, __riscv_vfncvt_x(v, vl), vl);
327#include <riscv_vector.h>
329static inline void volk_32fc_s32f_magnitude_16i_rvvseg(int16_t* magnitudeVector,
332 unsigned int num_points)
334 size_t n = num_points;
335 for (
size_t vl; n > 0; n -= vl, complexVector += vl, magnitudeVector += vl) {
336 vl = __riscv_vsetvl_e32m4(n);
337 vfloat32m4x2_t vc = __riscv_vlseg2e32_v_f32m4x2((
const float*)complexVector, vl);
338 vfloat32m4_t vr = __riscv_vget_f32m4(vc, 0);
339 vfloat32m4_t vi = __riscv_vget_f32m4(vc, 1);
340 vfloat32m4_t v = __riscv_vfmacc(__riscv_vfmul(vi, vi, vl), vr, vr, vl);
341 v = __riscv_vfmul(__riscv_vfsqrt(v, vl), scalar, vl);
342 __riscv_vse16(magnitudeVector, __riscv_vfncvt_x(v, vl), vl);
static float rintf(float x)
Definition config.h:45
static void volk_32fc_s32f_magnitude_16i_generic(int16_t *magnitudeVector, const lv_32fc_t *complexVector, const float scalar, unsigned int num_points)
Definition volk_32fc_s32f_magnitude_16i.h:70
static void volk_32fc_s32f_magnitude_16i_a_sse(int16_t *magnitudeVector, const lv_32fc_t *complexVector, const float scalar, unsigned int num_points)
Definition volk_32fc_s32f_magnitude_16i.h:191
static void volk_32fc_s32f_magnitude_16i_a_sse3(int16_t *magnitudeVector, const lv_32fc_t *complexVector, const float scalar, unsigned int num_points)
Definition volk_32fc_s32f_magnitude_16i.h:141
#define __VOLK_ATTR_ALIGNED(x)
Definition volk_common.h:62
float complex lv_32fc_t
Definition volk_complex.h:74