43#ifndef INCLUDED_volk_16ic_s32f_deinterleave_32f_x2_a_H
44#define INCLUDED_volk_16ic_s32f_deinterleave_32f_x2_a_H
54volk_16ic_s32f_deinterleave_32f_x2_a_avx2(
float* iBuffer,
58 unsigned int num_points)
60 float* iBufferPtr = iBuffer;
61 float* qBufferPtr = qBuffer;
64 const uint64_t eighthPoints = num_points / 8;
65 __m256 cplxValue1, cplxValue2, iValue, qValue;
66 __m256i cplxValueA, cplxValueB;
69 __m256 invScalar = _mm256_set1_ps(1.0 / scalar);
70 int16_t* complexVectorPtr = (int16_t*)complexVector;
71 __m256i idx = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0);
73 for (; number < eighthPoints; number++) {
75 cplxValueA = _mm256_load_si256((__m256i*)complexVectorPtr);
76 complexVectorPtr += 16;
79 cplxValue128 = _mm256_extracti128_si256(cplxValueA, 0);
80 cplxValueB = _mm256_cvtepi16_epi32(cplxValue128);
81 cplxValue1 = _mm256_cvtepi32_ps(cplxValueB);
82 cplxValue128 = _mm256_extracti128_si256(cplxValueA, 1);
83 cplxValueB = _mm256_cvtepi16_epi32(cplxValue128);
84 cplxValue2 = _mm256_cvtepi32_ps(cplxValueB);
86 cplxValue1 = _mm256_mul_ps(cplxValue1, invScalar);
87 cplxValue2 = _mm256_mul_ps(cplxValue2, invScalar);
90 iValue = _mm256_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(2, 0, 2, 0));
91 iValue = _mm256_permutevar8x32_ps(iValue, idx);
93 qValue = _mm256_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(3, 1, 3, 1));
94 qValue = _mm256_permutevar8x32_ps(qValue, idx);
96 _mm256_store_ps(iBufferPtr, iValue);
97 _mm256_store_ps(qBufferPtr, qValue);
103 number = eighthPoints * 8;
104 complexVectorPtr = (int16_t*)&complexVector[number];
105 for (; number < num_points; number++) {
106 *iBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
107 *qBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
113#include <xmmintrin.h>
120 unsigned int num_points)
122 float* iBufferPtr = iBuffer;
123 float* qBufferPtr = qBuffer;
126 const uint64_t quarterPoints = num_points / 4;
127 __m128 cplxValue1, cplxValue2, iValue, qValue;
129 __m128 invScalar = _mm_set_ps1(1.0 / scalar);
130 int16_t* complexVectorPtr = (int16_t*)complexVector;
134 for (; number < quarterPoints; number++) {
136 floatBuffer[0] = (float)(complexVectorPtr[0]);
137 floatBuffer[1] = (float)(complexVectorPtr[1]);
138 floatBuffer[2] = (float)(complexVectorPtr[2]);
139 floatBuffer[3] = (float)(complexVectorPtr[3]);
141 floatBuffer[4] = (float)(complexVectorPtr[4]);
142 floatBuffer[5] = (float)(complexVectorPtr[5]);
143 floatBuffer[6] = (float)(complexVectorPtr[6]);
144 floatBuffer[7] = (float)(complexVectorPtr[7]);
146 cplxValue1 = _mm_load_ps(&floatBuffer[0]);
147 cplxValue2 = _mm_load_ps(&floatBuffer[4]);
149 complexVectorPtr += 8;
151 cplxValue1 = _mm_mul_ps(cplxValue1, invScalar);
152 cplxValue2 = _mm_mul_ps(cplxValue2, invScalar);
155 iValue = _mm_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(2, 0, 2, 0));
157 qValue = _mm_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(3, 1, 3, 1));
159 _mm_store_ps(iBufferPtr, iValue);
160 _mm_store_ps(qBufferPtr, qValue);
166 number = quarterPoints * 4;
167 complexVectorPtr = (int16_t*)&complexVector[number];
168 for (; number < num_points; number++) {
169 *iBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
170 *qBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
175#ifdef LV_HAVE_GENERIC
182 unsigned int num_points)
184 const int16_t* complexVectorPtr = (
const int16_t*)complexVector;
185 float* iBufferPtr = iBuffer;
186 float* qBufferPtr = qBuffer;
188 for (number = 0; number < num_points; number++) {
189 *iBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
190 *qBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
201 unsigned int num_points)
203 const int16_t* complexVectorPtr = (
const int16_t*)complexVector;
204 float* iBufferPtr = iBuffer;
205 float* qBufferPtr = qBuffer;
206 unsigned int eighth_points = num_points / 4;
208 float iScalar = 1.f / scalar;
209 float32x4_t invScalar;
210 invScalar = vld1q_dup_f32(&iScalar);
212 int16x4x2_t complexInput_s16;
213 int32x4x2_t complexInput_s32;
214 float32x4x2_t complexFloat;
216 for (number = 0; number < eighth_points; number++) {
217 complexInput_s16 = vld2_s16(complexVectorPtr);
218 complexInput_s32.val[0] = vmovl_s16(complexInput_s16.val[0]);
219 complexInput_s32.val[1] = vmovl_s16(complexInput_s16.val[1]);
220 complexFloat.val[0] = vcvtq_f32_s32(complexInput_s32.val[0]);
221 complexFloat.val[1] = vcvtq_f32_s32(complexInput_s32.val[1]);
222 complexFloat.val[0] = vmulq_f32(complexFloat.val[0], invScalar);
223 complexFloat.val[1] = vmulq_f32(complexFloat.val[1], invScalar);
224 vst1q_f32(iBufferPtr, complexFloat.val[0]);
225 vst1q_f32(qBufferPtr, complexFloat.val[1]);
226 complexVectorPtr += 8;
231 for (number = eighth_points * 4; number < num_points; number++) {
232 *iBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
233 *qBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
239extern void volk_16ic_s32f_deinterleave_32f_x2_a_orc_impl(
float* iBuffer,
246volk_16ic_s32f_deinterleave_32f_x2_u_orc(
float* iBuffer,
250 unsigned int num_points)
252 volk_16ic_s32f_deinterleave_32f_x2_a_orc_impl(
253 iBuffer, qBuffer, complexVector, scalar, num_points);
261#ifndef INCLUDED_volk_16ic_s32f_deinterleave_32f_x2_u_H
262#define INCLUDED_volk_16ic_s32f_deinterleave_32f_x2_u_H
269#include <immintrin.h>
272volk_16ic_s32f_deinterleave_32f_x2_u_avx2(
float* iBuffer,
276 unsigned int num_points)
278 float* iBufferPtr = iBuffer;
279 float* qBufferPtr = qBuffer;
282 const uint64_t eighthPoints = num_points / 8;
283 __m256 cplxValue1, cplxValue2, iValue, qValue;
284 __m256i cplxValueA, cplxValueB;
285 __m128i cplxValue128;
287 __m256 invScalar = _mm256_set1_ps(1.0 / scalar);
288 int16_t* complexVectorPtr = (int16_t*)complexVector;
289 __m256i idx = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0);
291 for (; number < eighthPoints; number++) {
293 cplxValueA = _mm256_loadu_si256((__m256i*)complexVectorPtr);
294 complexVectorPtr += 16;
297 cplxValue128 = _mm256_extracti128_si256(cplxValueA, 0);
298 cplxValueB = _mm256_cvtepi16_epi32(cplxValue128);
299 cplxValue1 = _mm256_cvtepi32_ps(cplxValueB);
300 cplxValue128 = _mm256_extracti128_si256(cplxValueA, 1);
301 cplxValueB = _mm256_cvtepi16_epi32(cplxValue128);
302 cplxValue2 = _mm256_cvtepi32_ps(cplxValueB);
304 cplxValue1 = _mm256_mul_ps(cplxValue1, invScalar);
305 cplxValue2 = _mm256_mul_ps(cplxValue2, invScalar);
308 iValue = _mm256_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(2, 0, 2, 0));
309 iValue = _mm256_permutevar8x32_ps(iValue, idx);
311 qValue = _mm256_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(3, 1, 3, 1));
312 qValue = _mm256_permutevar8x32_ps(qValue, idx);
314 _mm256_storeu_ps(iBufferPtr, iValue);
315 _mm256_storeu_ps(qBufferPtr, qValue);
321 number = eighthPoints * 8;
322 complexVectorPtr = (int16_t*)&complexVector[number];
323 for (; number < num_points; number++) {
324 *iBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
325 *qBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
331#include <riscv_vector.h>
333static inline void volk_16ic_s32f_deinterleave_32f_x2_rvv(
float* iBuffer,
337 unsigned int num_points)
339 size_t n = num_points;
340 for (
size_t vl; n > 0; n -= vl, complexVector += vl, iBuffer += vl, qBuffer += vl) {
341 vl = __riscv_vsetvl_e16m4(n);
342 vint32m8_t vc = __riscv_vle32_v_i32m8((
const int32_t*)complexVector, vl);
343 vint16m4_t vr = __riscv_vnsra(vc, 0, vl);
344 vint16m4_t vi = __riscv_vnsra(vc, 16, vl);
345 vfloat32m8_t vrf = __riscv_vfwcvt_f(vr, vl);
346 vfloat32m8_t vif = __riscv_vfwcvt_f(vi, vl);
347 __riscv_vse32(iBuffer, __riscv_vfmul(vrf, 1.0f / scalar, vl), vl);
348 __riscv_vse32(qBuffer, __riscv_vfmul(vif, 1.0f / scalar, vl), vl);
354#include <riscv_vector.h>
357volk_16ic_s32f_deinterleave_32f_x2_rvvseg(
float* iBuffer,
361 unsigned int num_points)
363 size_t n = num_points;
364 for (
size_t vl; n > 0; n -= vl, complexVector += vl, iBuffer += vl, qBuffer += vl) {
365 vl = __riscv_vsetvl_e16m4(n);
366 vint16m4x2_t vc = __riscv_vlseg2e16_v_i16m4x2((
const int16_t*)complexVector, vl);
367 vint16m4_t vr = __riscv_vget_i16m4(vc, 0);
368 vint16m4_t vi = __riscv_vget_i16m4(vc, 1);
369 vfloat32m8_t vrf = __riscv_vfwcvt_f(vr, vl);
370 vfloat32m8_t vif = __riscv_vfwcvt_f(vi, vl);
371 __riscv_vse32(iBuffer, __riscv_vfmul(vrf, 1.0f / scalar, vl), vl);
372 __riscv_vse32(qBuffer, __riscv_vfmul(vif, 1.0f / scalar, vl), vl);
static void volk_16ic_s32f_deinterleave_32f_x2_neon(float *iBuffer, float *qBuffer, const lv_16sc_t *complexVector, const float scalar, unsigned int num_points)
Definition volk_16ic_s32f_deinterleave_32f_x2.h:197
static void volk_16ic_s32f_deinterleave_32f_x2_generic(float *iBuffer, float *qBuffer, const lv_16sc_t *complexVector, const float scalar, unsigned int num_points)
Definition volk_16ic_s32f_deinterleave_32f_x2.h:178
static void volk_16ic_s32f_deinterleave_32f_x2_a_sse(float *iBuffer, float *qBuffer, const lv_16sc_t *complexVector, const float scalar, unsigned int num_points)
Definition volk_16ic_s32f_deinterleave_32f_x2.h:116
#define __VOLK_ATTR_ALIGNED(x)
Definition volk_common.h:62
short complex lv_16sc_t
Definition volk_complex.h:71