40#ifndef INCLUDED_volk_16i_convert_8i_u_H
41#define INCLUDED_volk_16i_convert_8i_u_H
49static inline void volk_16i_convert_8i_u_avx2(int8_t* outputVector,
50 const int16_t* inputVector,
51 unsigned int num_points)
53 unsigned int number = 0;
54 const unsigned int thirtysecondPoints = num_points / 32;
56 int8_t* outputVectorPtr = outputVector;
57 int16_t* inputPtr = (int16_t*)inputVector;
62 for (; number < thirtysecondPoints; number++) {
65 inputVal1 = _mm256_loadu_si256((__m256i*)inputPtr);
67 inputVal2 = _mm256_loadu_si256((__m256i*)inputPtr);
70 inputVal1 = _mm256_srai_epi16(inputVal1, 8);
71 inputVal2 = _mm256_srai_epi16(inputVal2, 8);
73 ret = _mm256_packs_epi16(inputVal1, inputVal2);
74 ret = _mm256_permute4x64_epi64(ret, 0b11011000);
76 _mm256_storeu_si256((__m256i*)outputVectorPtr, ret);
78 outputVectorPtr += 32;
81 number = thirtysecondPoints * 32;
82 for (; number < num_points; number++) {
83 outputVector[number] = (int8_t)(inputVector[number] >> 8);
93 const int16_t* inputVector,
94 unsigned int num_points)
96 unsigned int number = 0;
97 const unsigned int sixteenthPoints = num_points / 16;
99 int8_t* outputVectorPtr = outputVector;
100 int16_t* inputPtr = (int16_t*)inputVector;
105 for (; number < sixteenthPoints; number++) {
120 outputVectorPtr += 16;
123 number = sixteenthPoints * 16;
124 for (; number < num_points; number++) {
125 outputVector[number] = (int8_t)(inputVector[number] >> 8);
131#ifdef LV_HAVE_GENERIC
134 const int16_t* inputVector,
135 unsigned int num_points)
137 int8_t* outputVectorPtr = outputVector;
138 const int16_t* inputVectorPtr = inputVector;
139 unsigned int number = 0;
141 for (number = 0; number < num_points; number++) {
142 *outputVectorPtr++ = ((int8_t)(*inputVectorPtr++ >> 8));
149#ifndef INCLUDED_volk_16i_convert_8i_a_H
150#define INCLUDED_volk_16i_convert_8i_a_H
156#include <immintrin.h>
158static inline void volk_16i_convert_8i_a_avx2(int8_t* outputVector,
159 const int16_t* inputVector,
160 unsigned int num_points)
162 unsigned int number = 0;
163 const unsigned int thirtysecondPoints = num_points / 32;
165 int8_t* outputVectorPtr = outputVector;
166 int16_t* inputPtr = (int16_t*)inputVector;
171 for (; number < thirtysecondPoints; number++) {
174 inputVal1 = _mm256_load_si256((__m256i*)inputPtr);
176 inputVal2 = _mm256_load_si256((__m256i*)inputPtr);
179 inputVal1 = _mm256_srai_epi16(inputVal1, 8);
180 inputVal2 = _mm256_srai_epi16(inputVal2, 8);
182 ret = _mm256_packs_epi16(inputVal1, inputVal2);
183 ret = _mm256_permute4x64_epi64(ret, 0b11011000);
185 _mm256_store_si256((__m256i*)outputVectorPtr, ret);
187 outputVectorPtr += 32;
190 number = thirtysecondPoints * 32;
191 for (; number < num_points; number++) {
192 outputVector[number] = (int8_t)(inputVector[number] >> 8);
199#include <emmintrin.h>
202 const int16_t* inputVector,
203 unsigned int num_points)
205 unsigned int number = 0;
206 const unsigned int sixteenthPoints = num_points / 16;
208 int8_t* outputVectorPtr = outputVector;
209 int16_t* inputPtr = (int16_t*)inputVector;
214 for (; number < sixteenthPoints; number++) {
229 outputVectorPtr += 16;
232 number = sixteenthPoints * 16;
233 for (; number < num_points; number++) {
234 outputVector[number] = (int8_t)(inputVector[number] >> 8);
244 const int16_t* inputVector,
245 unsigned int num_points)
247 int8_t* outputVectorPtr = outputVector;
248 const int16_t* inputVectorPtr = inputVector;
249 unsigned int number = 0;
250 unsigned int sixteenth_points = num_points / 16;
258 for (number = 0; number < sixteenth_points; number++) {
260 inputVal0 = vld1q_s16(inputVectorPtr);
261 inputVal1 = vld1q_s16(inputVectorPtr + 8);
263 outputVal0 = vshrn_n_s16(inputVal0, 8);
264 outputVal1 = vshrn_n_s16(inputVal1, 8);
266 outputVal = vcombine_s8(outputVal0, outputVal1);
267 vst1q_s8(outputVectorPtr, outputVal);
268 inputVectorPtr += 16;
269 outputVectorPtr += 16;
272 for (number = sixteenth_points * 16; number < num_points; number++) {
273 *outputVectorPtr++ = ((int8_t)(*inputVectorPtr++ >> 8));