56 #ifndef INCLUDED_volk_16ic_s32f_deinterleave_32f_x2_a_H
57 #define INCLUDED_volk_16ic_s32f_deinterleave_32f_x2_a_H
64 #include <immintrin.h>
67 volk_16ic_s32f_deinterleave_32f_x2_a_avx2(
float* iBuffer,
71 unsigned int num_points)
73 float* iBufferPtr = iBuffer;
74 float* qBufferPtr = qBuffer;
77 const uint64_t eighthPoints = num_points / 8;
78 __m256 cplxValue1, cplxValue2, iValue, qValue;
79 __m256i cplxValueA, cplxValueB;
82 __m256 invScalar = _mm256_set1_ps(1.0 / scalar);
83 int16_t* complexVectorPtr = (int16_t*)complexVector;
84 __m256i idx = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0);
86 for (; number < eighthPoints; number++) {
88 cplxValueA = _mm256_load_si256((__m256i*)complexVectorPtr);
89 complexVectorPtr += 16;
92 cplxValue128 = _mm256_extracti128_si256(cplxValueA, 0);
93 cplxValueB = _mm256_cvtepi16_epi32(cplxValue128);
94 cplxValue1 = _mm256_cvtepi32_ps(cplxValueB);
95 cplxValue128 = _mm256_extracti128_si256(cplxValueA, 1);
96 cplxValueB = _mm256_cvtepi16_epi32(cplxValue128);
97 cplxValue2 = _mm256_cvtepi32_ps(cplxValueB);
99 cplxValue1 = _mm256_mul_ps(cplxValue1, invScalar);
100 cplxValue2 = _mm256_mul_ps(cplxValue2, invScalar);
103 iValue = _mm256_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(2, 0, 2, 0));
104 iValue = _mm256_permutevar8x32_ps(iValue, idx);
106 qValue = _mm256_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(3, 1, 3, 1));
107 qValue = _mm256_permutevar8x32_ps(qValue, idx);
109 _mm256_store_ps(iBufferPtr, iValue);
110 _mm256_store_ps(qBufferPtr, qValue);
116 number = eighthPoints * 8;
117 complexVectorPtr = (int16_t*)&complexVector[number];
118 for (; number < num_points; number++) {
119 *iBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
120 *qBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
126 #include <xmmintrin.h>
133 unsigned int num_points)
135 float* iBufferPtr = iBuffer;
136 float* qBufferPtr = qBuffer;
139 const uint64_t quarterPoints = num_points / 4;
140 __m128 cplxValue1, cplxValue2, iValue, qValue;
142 __m128 invScalar = _mm_set_ps1(1.0 / scalar);
143 int16_t* complexVectorPtr = (int16_t*)complexVector;
147 for (; number < quarterPoints; number++) {
149 floatBuffer[0] = (float)(complexVectorPtr[0]);
150 floatBuffer[1] = (float)(complexVectorPtr[1]);
151 floatBuffer[2] = (float)(complexVectorPtr[2]);
152 floatBuffer[3] = (float)(complexVectorPtr[3]);
154 floatBuffer[4] = (float)(complexVectorPtr[4]);
155 floatBuffer[5] = (float)(complexVectorPtr[5]);
156 floatBuffer[6] = (float)(complexVectorPtr[6]);
157 floatBuffer[7] = (float)(complexVectorPtr[7]);
159 cplxValue1 = _mm_load_ps(&floatBuffer[0]);
160 cplxValue2 = _mm_load_ps(&floatBuffer[4]);
162 complexVectorPtr += 8;
164 cplxValue1 = _mm_mul_ps(cplxValue1, invScalar);
165 cplxValue2 = _mm_mul_ps(cplxValue2, invScalar);
168 iValue = _mm_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(2, 0, 2, 0));
170 qValue = _mm_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(3, 1, 3, 1));
172 _mm_store_ps(iBufferPtr, iValue);
173 _mm_store_ps(qBufferPtr, qValue);
179 number = quarterPoints * 4;
180 complexVectorPtr = (int16_t*)&complexVector[number];
181 for (; number < num_points; number++) {
182 *iBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
183 *qBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
188 #ifdef LV_HAVE_GENERIC
195 unsigned int num_points)
197 const int16_t* complexVectorPtr = (
const int16_t*)complexVector;
198 float* iBufferPtr = iBuffer;
199 float* qBufferPtr = qBuffer;
201 for (number = 0; number < num_points; number++) {
202 *iBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
203 *qBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
209 #include <arm_neon.h>
214 unsigned int num_points)
216 const int16_t* complexVectorPtr = (
const int16_t*)complexVector;
217 float* iBufferPtr = iBuffer;
218 float* qBufferPtr = qBuffer;
219 unsigned int eighth_points = num_points / 4;
221 float iScalar = 1.f / scalar;
222 float32x4_t invScalar;
223 invScalar = vld1q_dup_f32(&iScalar);
225 int16x4x2_t complexInput_s16;
226 int32x4x2_t complexInput_s32;
227 float32x4x2_t complexFloat;
229 for (number = 0; number < eighth_points; number++) {
230 complexInput_s16 = vld2_s16(complexVectorPtr);
231 complexInput_s32.val[0] = vmovl_s16(complexInput_s16.val[0]);
232 complexInput_s32.val[1] = vmovl_s16(complexInput_s16.val[1]);
233 complexFloat.val[0] = vcvtq_f32_s32(complexInput_s32.val[0]);
234 complexFloat.val[1] = vcvtq_f32_s32(complexInput_s32.val[1]);
235 complexFloat.val[0] = vmulq_f32(complexFloat.val[0], invScalar);
236 complexFloat.val[1] = vmulq_f32(complexFloat.val[1], invScalar);
237 vst1q_f32(iBufferPtr, complexFloat.val[0]);
238 vst1q_f32(qBufferPtr, complexFloat.val[1]);
239 complexVectorPtr += 8;
244 for (number = eighth_points * 4; number < num_points; number++) {
245 *iBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
246 *qBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
252 extern void volk_16ic_s32f_deinterleave_32f_x2_a_orc_impl(
float* iBuffer,
256 unsigned int num_points);
259 volk_16ic_s32f_deinterleave_32f_x2_u_orc(
float* iBuffer,
263 unsigned int num_points)
265 volk_16ic_s32f_deinterleave_32f_x2_a_orc_impl(
266 iBuffer, qBuffer, complexVector, scalar, num_points);
274 #ifndef INCLUDED_volk_16ic_s32f_deinterleave_32f_x2_u_H
275 #define INCLUDED_volk_16ic_s32f_deinterleave_32f_x2_u_H
277 #include <inttypes.h>
282 #include <immintrin.h>
285 volk_16ic_s32f_deinterleave_32f_x2_u_avx2(
float* iBuffer,
289 unsigned int num_points)
291 float* iBufferPtr = iBuffer;
292 float* qBufferPtr = qBuffer;
295 const uint64_t eighthPoints = num_points / 8;
296 __m256 cplxValue1, cplxValue2, iValue, qValue;
297 __m256i cplxValueA, cplxValueB;
298 __m128i cplxValue128;
300 __m256 invScalar = _mm256_set1_ps(1.0 / scalar);
301 int16_t* complexVectorPtr = (int16_t*)complexVector;
302 __m256i idx = _mm256_set_epi32(7, 6, 3, 2, 5, 4, 1, 0);
304 for (; number < eighthPoints; number++) {
306 cplxValueA = _mm256_loadu_si256((__m256i*)complexVectorPtr);
307 complexVectorPtr += 16;
310 cplxValue128 = _mm256_extracti128_si256(cplxValueA, 0);
311 cplxValueB = _mm256_cvtepi16_epi32(cplxValue128);
312 cplxValue1 = _mm256_cvtepi32_ps(cplxValueB);
313 cplxValue128 = _mm256_extracti128_si256(cplxValueA, 1);
314 cplxValueB = _mm256_cvtepi16_epi32(cplxValue128);
315 cplxValue2 = _mm256_cvtepi32_ps(cplxValueB);
317 cplxValue1 = _mm256_mul_ps(cplxValue1, invScalar);
318 cplxValue2 = _mm256_mul_ps(cplxValue2, invScalar);
321 iValue = _mm256_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(2, 0, 2, 0));
322 iValue = _mm256_permutevar8x32_ps(iValue, idx);
324 qValue = _mm256_shuffle_ps(cplxValue1, cplxValue2, _MM_SHUFFLE(3, 1, 3, 1));
325 qValue = _mm256_permutevar8x32_ps(qValue, idx);
327 _mm256_storeu_ps(iBufferPtr, iValue);
328 _mm256_storeu_ps(qBufferPtr, qValue);
334 number = eighthPoints * 8;
335 complexVectorPtr = (int16_t*)&complexVector[number];
336 for (; number < num_points; number++) {
337 *iBufferPtr++ = (float)(*complexVectorPtr++) / scalar;
338 *qBufferPtr++ = (float)(*complexVectorPtr++) / scalar;