42#ifndef INCLUDED_volk_16ic_s32f_magnitude_32f_a_H
43#define INCLUDED_volk_16ic_s32f_magnitude_32f_a_H
53static inline void volk_16ic_s32f_magnitude_32f_a_avx2(
float* magnitudeVector,
56 unsigned int num_points)
58 unsigned int number = 0;
59 const unsigned int eighthPoints = num_points / 8;
61 const int16_t* complexVectorPtr = (
const int16_t*)complexVector;
62 float* magnitudeVectorPtr = magnitudeVector;
64 __m256 invScalar = _mm256_set1_ps(1.0 / scalar);
66 __m256 cplxValue1, cplxValue2, result;
68 __m128i short1, short2;
69 __m256i idx = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0);
71 for (; number < eighthPoints; number++) {
73 int1 = _mm256_loadu_si256((__m256i*)complexVectorPtr);
74 complexVectorPtr += 16;
75 short1 = _mm256_extracti128_si256(int1, 0);
76 short2 = _mm256_extracti128_si256(int1, 1);
78 int1 = _mm256_cvtepi16_epi32(short1);
79 int2 = _mm256_cvtepi16_epi32(short2);
80 cplxValue1 = _mm256_cvtepi32_ps(int1);
81 cplxValue2 = _mm256_cvtepi32_ps(int2);
83 cplxValue1 = _mm256_mul_ps(cplxValue1, invScalar);
84 cplxValue2 = _mm256_mul_ps(cplxValue2, invScalar);
86 cplxValue1 = _mm256_mul_ps(cplxValue1, cplxValue1);
87 cplxValue2 = _mm256_mul_ps(cplxValue2, cplxValue2);
89 result = _mm256_hadd_ps(cplxValue1, cplxValue2);
90 result = _mm256_permutevar8x32_ps(result, idx);
92 result = _mm256_sqrt_ps(result);
94 _mm256_store_ps(magnitudeVectorPtr, result);
96 magnitudeVectorPtr += 8;
99 number = eighthPoints * 8;
100 magnitudeVectorPtr = &magnitudeVector[number];
101 complexVectorPtr = (
const int16_t*)&complexVector[number];
102 for (; number < num_points; number++) {
103 float val1Real = (float)(*complexVectorPtr++) / scalar;
104 float val1Imag = (float)(*complexVectorPtr++) / scalar;
105 *magnitudeVectorPtr++ = sqrtf((val1Real * val1Real) + (val1Imag * val1Imag));
112#include <pmmintrin.h>
117 unsigned int num_points)
119 unsigned int number = 0;
120 const unsigned int quarterPoints = num_points / 4;
122 const int16_t* complexVectorPtr = (
const int16_t*)complexVector;
123 float* magnitudeVectorPtr = magnitudeVector;
125 __m128 invScalar = _mm_set_ps1(1.0 / scalar);
127 __m128 cplxValue1, cplxValue2, result;
131 for (; number < quarterPoints; number++) {
133 inputFloatBuffer[0] = (float)(complexVectorPtr[0]);
134 inputFloatBuffer[1] = (float)(complexVectorPtr[1]);
135 inputFloatBuffer[2] = (float)(complexVectorPtr[2]);
136 inputFloatBuffer[3] = (float)(complexVectorPtr[3]);
138 inputFloatBuffer[4] = (float)(complexVectorPtr[4]);
139 inputFloatBuffer[5] = (float)(complexVectorPtr[5]);
140 inputFloatBuffer[6] = (float)(complexVectorPtr[6]);
141 inputFloatBuffer[7] = (float)(complexVectorPtr[7]);
143 cplxValue1 = _mm_load_ps(&inputFloatBuffer[0]);
144 cplxValue2 = _mm_load_ps(&inputFloatBuffer[4]);
146 complexVectorPtr += 8;
148 cplxValue1 = _mm_mul_ps(cplxValue1, invScalar);
149 cplxValue2 = _mm_mul_ps(cplxValue2, invScalar);
151 cplxValue1 = _mm_mul_ps(cplxValue1, cplxValue1);
152 cplxValue2 = _mm_mul_ps(cplxValue2, cplxValue2);
154 result = _mm_hadd_ps(cplxValue1, cplxValue2);
156 result = _mm_sqrt_ps(result);
158 _mm_store_ps(magnitudeVectorPtr, result);
160 magnitudeVectorPtr += 4;
163 number = quarterPoints * 4;
164 magnitudeVectorPtr = &magnitudeVector[number];
165 complexVectorPtr = (
const int16_t*)&complexVector[number];
166 for (; number < num_points; number++) {
167 float val1Real = (float)(*complexVectorPtr++) / scalar;
168 float val1Imag = (float)(*complexVectorPtr++) / scalar;
169 *magnitudeVectorPtr++ = sqrtf((val1Real * val1Real) + (val1Imag * val1Imag));
175#include <xmmintrin.h>
180 unsigned int num_points)
182 unsigned int number = 0;
183 const unsigned int quarterPoints = num_points / 4;
185 const int16_t* complexVectorPtr = (
const int16_t*)complexVector;
186 float* magnitudeVectorPtr = magnitudeVector;
188 const float iScalar = 1.0 / scalar;
189 __m128 invScalar = _mm_set_ps1(iScalar);
191 __m128 cplxValue1, cplxValue2, result, re, im;
195 for (; number < quarterPoints; number++) {
196 inputFloatBuffer[0] = (float)(complexVectorPtr[0]);
197 inputFloatBuffer[1] = (float)(complexVectorPtr[1]);
198 inputFloatBuffer[2] = (float)(complexVectorPtr[2]);
199 inputFloatBuffer[3] = (float)(complexVectorPtr[3]);
201 inputFloatBuffer[4] = (float)(complexVectorPtr[4]);
202 inputFloatBuffer[5] = (float)(complexVectorPtr[5]);
203 inputFloatBuffer[6] = (float)(complexVectorPtr[6]);
204 inputFloatBuffer[7] = (float)(complexVectorPtr[7]);
206 cplxValue1 = _mm_load_ps(&inputFloatBuffer[0]);
207 cplxValue2 = _mm_load_ps(&inputFloatBuffer[4]);
209 re = _mm_shuffle_ps(cplxValue1, cplxValue2, 0x88);
210 im = _mm_shuffle_ps(cplxValue1, cplxValue2, 0xdd);
212 complexVectorPtr += 8;
214 cplxValue1 = _mm_mul_ps(re, invScalar);
215 cplxValue2 = _mm_mul_ps(im, invScalar);
217 cplxValue1 = _mm_mul_ps(cplxValue1, cplxValue1);
218 cplxValue2 = _mm_mul_ps(cplxValue2, cplxValue2);
220 result = _mm_add_ps(cplxValue1, cplxValue2);
222 result = _mm_sqrt_ps(result);
224 _mm_store_ps(magnitudeVectorPtr, result);
226 magnitudeVectorPtr += 4;
229 number = quarterPoints * 4;
230 magnitudeVectorPtr = &magnitudeVector[number];
231 complexVectorPtr = (
const int16_t*)&complexVector[number];
232 for (; number < num_points; number++) {
233 float val1Real = (float)(*complexVectorPtr++) * iScalar;
234 float val1Imag = (float)(*complexVectorPtr++) * iScalar;
235 *magnitudeVectorPtr++ = sqrtf((val1Real * val1Real) + (val1Imag * val1Imag));
242#ifdef LV_HAVE_GENERIC
247 unsigned int num_points)
249 const int16_t* complexVectorPtr = (
const int16_t*)complexVector;
250 float* magnitudeVectorPtr = magnitudeVector;
251 unsigned int number = 0;
252 const float invScalar = 1.0 / scalar;
253 for (number = 0; number < num_points; number++) {
254 float real = ((float)(*complexVectorPtr++)) * invScalar;
255 float imag = ((float)(*complexVectorPtr++)) * invScalar;
256 *magnitudeVectorPtr++ = sqrtf((real * real) + (imag * imag));
264#ifndef INCLUDED_volk_16ic_s32f_magnitude_32f_u_H
265#define INCLUDED_volk_16ic_s32f_magnitude_32f_u_H
273#include <immintrin.h>
275static inline void volk_16ic_s32f_magnitude_32f_u_avx2(
float* magnitudeVector,
278 unsigned int num_points)
280 unsigned int number = 0;
281 const unsigned int eighthPoints = num_points / 8;
283 const int16_t* complexVectorPtr = (
const int16_t*)complexVector;
284 float* magnitudeVectorPtr = magnitudeVector;
286 __m256 invScalar = _mm256_set1_ps(1.0 / scalar);
288 __m256 cplxValue1, cplxValue2, result;
290 __m128i short1, short2;
291 __m256i idx = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0);
293 for (; number < eighthPoints; number++) {
295 int1 = _mm256_loadu_si256((__m256i*)complexVectorPtr);
296 complexVectorPtr += 16;
297 short1 = _mm256_extracti128_si256(int1, 0);
298 short2 = _mm256_extracti128_si256(int1, 1);
300 int1 = _mm256_cvtepi16_epi32(short1);
301 int2 = _mm256_cvtepi16_epi32(short2);
302 cplxValue1 = _mm256_cvtepi32_ps(int1);
303 cplxValue2 = _mm256_cvtepi32_ps(int2);
305 cplxValue1 = _mm256_mul_ps(cplxValue1, invScalar);
306 cplxValue2 = _mm256_mul_ps(cplxValue2, invScalar);
308 cplxValue1 = _mm256_mul_ps(cplxValue1, cplxValue1);
309 cplxValue2 = _mm256_mul_ps(cplxValue2, cplxValue2);
311 result = _mm256_hadd_ps(cplxValue1, cplxValue2);
312 result = _mm256_permutevar8x32_ps(result, idx);
314 result = _mm256_sqrt_ps(result);
316 _mm256_storeu_ps(magnitudeVectorPtr, result);
318 magnitudeVectorPtr += 8;
321 number = eighthPoints * 8;
322 magnitudeVectorPtr = &magnitudeVector[number];
323 complexVectorPtr = (
const int16_t*)&complexVector[number];
324 for (; number < num_points; number++) {
325 float val1Real = (float)(*complexVectorPtr++) / scalar;
326 float val1Imag = (float)(*complexVectorPtr++) / scalar;
327 *magnitudeVectorPtr++ = sqrtf((val1Real * val1Real) + (val1Imag * val1Imag));
333#include <riscv_vector.h>
335static inline void volk_16ic_s32f_magnitude_32f_rvv(
float* magnitudeVector,
338 unsigned int num_points)
340 size_t n = num_points;
341 for (
size_t vl; n > 0; n -= vl, complexVector += vl, magnitudeVector += vl) {
342 vl = __riscv_vsetvl_e16m4(n);
343 vint32m8_t vc = __riscv_vle32_v_i32m8((
const int32_t*)complexVector, vl);
344 vint16m4_t vr = __riscv_vnsra(vc, 0, vl);
345 vint16m4_t vi = __riscv_vnsra(vc, 16, vl);
346 vfloat32m8_t vrf = __riscv_vfmul(__riscv_vfwcvt_f(vr, vl), 1.0f / scalar, vl);
347 vfloat32m8_t vif = __riscv_vfmul(__riscv_vfwcvt_f(vi, vl), 1.0f / scalar, vl);
348 vfloat32m8_t vf = __riscv_vfmacc(__riscv_vfmul(vif, vif, vl), vrf, vrf, vl);
349 __riscv_vse32(magnitudeVector, __riscv_vfsqrt(vf, vl), vl);
355#include <riscv_vector.h>
357static inline void volk_16ic_s32f_magnitude_32f_rvvseg(
float* magnitudeVector,
360 unsigned int num_points)
362 size_t n = num_points;
363 for (
size_t vl; n > 0; n -= vl, complexVector += vl, magnitudeVector += vl) {
364 vl = __riscv_vsetvl_e16m4(n);
365 vint16m4x2_t vc = __riscv_vlseg2e16_v_i16m4x2((
const int16_t*)complexVector, vl);
366 vint16m4_t vr = __riscv_vget_i16m4(vc, 0);
367 vint16m4_t vi = __riscv_vget_i16m4(vc, 1);
368 vfloat32m8_t vrf = __riscv_vfmul(__riscv_vfwcvt_f(vr, vl), 1.0f / scalar, vl);
369 vfloat32m8_t vif = __riscv_vfmul(__riscv_vfwcvt_f(vi, vl), 1.0f / scalar, vl);
370 vfloat32m8_t vf = __riscv_vfmacc(__riscv_vfmul(vif, vif, vl), vrf, vrf, vl);
371 __riscv_vse32(magnitudeVector, __riscv_vfsqrt(vf, vl), vl);
static void volk_16ic_s32f_magnitude_32f_generic(float *magnitudeVector, const lv_16sc_t *complexVector, const float scalar, unsigned int num_points)
Definition volk_16ic_s32f_magnitude_32f.h:244
static void volk_16ic_s32f_magnitude_32f_a_sse(float *magnitudeVector, const lv_16sc_t *complexVector, const float scalar, unsigned int num_points)
Definition volk_16ic_s32f_magnitude_32f.h:177
static void volk_16ic_s32f_magnitude_32f_a_sse3(float *magnitudeVector, const lv_16sc_t *complexVector, const float scalar, unsigned int num_points)
Definition volk_16ic_s32f_magnitude_32f.h:114
#define __VOLK_ATTR_ALIGNED(x)
Definition volk_common.h:62
short complex lv_16sc_t
Definition volk_complex.h:71