mirrors/gnss-sdr
1
0
Fork 0
mirror of https://github.com/gnss-sdr/gnss-sdr synced 2024-06-18 02:59:58 +00:00
Code Issues Releases Wiki Activity

Add AVX implementation

Browse Source
This commit is contained in:
Carles Fernandez 2018-08-09 22:00:22 +02:00
parent c5f10cd56c
commit 66bfbffe89
No known key found for this signature in database
GPG Key ID: 4C583C52B0C3877D
2 changed files with 238 additions and 214 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

120
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_32f_fast_resamplerxnpuppet_32f.h
View File

@ -71,9 +71,9 @@ static inline void volk_gnsssdr_32f_fast_resamplerxnpuppet_32f_generic(float* re
volk_gnsssdr_free(result_aux);
}
#endif /* LV_HAVE_GENERIC */
#ifdef LV_HAVE_SSE3
static inline void volk_gnsssdr_32f_fast_resamplerxnpuppet_32f_a_sse3(float* result, const float* local_code, unsigned int num_points)
{
@ -104,6 +104,7 @@ static inline void volk_gnsssdr_32f_fast_resamplerxnpuppet_32f_a_sse3(float* res
#endif
#ifdef LV_HAVE_SSE3
static inline void volk_gnsssdr_32f_fast_resamplerxnpuppet_32f_u_sse3(float* result, const float* local_code, unsigned int num_points)
{
@ -195,63 +196,66 @@ static inline void volk_gnsssdr_32f_fast_resamplerxnpuppet_32f_a_sse4_1(float* r
}
#endif
//
//#ifdef LV_HAVE_AVX
//static inline void volk_gnsssdr_32f_resamplerxnpuppet_32f_a_avx(float* result, const float* local_code, unsigned int num_points)
//{
// int code_length_chips = 2046;
// float code_phase_step_chips = ((float)(code_length_chips) + 0.1) / ((float)num_points);
// int num_out_vectors = 3;
// float rem_code_phase_chips = -0.234;
// unsigned int n;
// float shifts_chips[3] = {-0.1, 0.0, 0.1};
//
// float** result_aux = (float**)volk_gnsssdr_malloc(sizeof(float*) * num_out_vectors, volk_gnsssdr_get_alignment());
// for (n = 0; n < num_out_vectors; n++)
// {
// result_aux[n] = (float*)volk_gnsssdr_malloc(sizeof(float) * num_points, volk_gnsssdr_get_alignment());
// }
//
// volk_gnsssdr_32f_xn_resampler_32f_xn_a_avx(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
//
// memcpy((float*)result, (float*)result_aux[0], sizeof(float) * num_points);
//
// for (n = 0; n < num_out_vectors; n++)
// {
// volk_gnsssdr_free(result_aux[n]);
// }
// volk_gnsssdr_free(result_aux);
//}
//#endif
//
//
//#ifdef LV_HAVE_AVX
//static inline void volk_gnsssdr_32f_resamplerxnpuppet_32f_u_avx(float* result, const float* local_code, unsigned int num_points)
//{
// int code_length_chips = 2046;
// float code_phase_step_chips = ((float)(code_length_chips) + 0.1) / ((float)num_points);
// int num_out_vectors = 3;
// float rem_code_phase_chips = -0.234;
// unsigned int n;
// float shifts_chips[3] = {-0.1, 0.0, 0.1};
//
// float** result_aux = (float**)volk_gnsssdr_malloc(sizeof(float*) * num_out_vectors, volk_gnsssdr_get_alignment());
// for (n = 0; n < num_out_vectors; n++)
// {
// result_aux[n] = (float*)volk_gnsssdr_malloc(sizeof(float) * num_points, volk_gnsssdr_get_alignment());
// }
//
// volk_gnsssdr_32f_xn_resampler_32f_xn_u_avx(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
//
// memcpy((float*)result, (float*)result_aux[0], sizeof(float) * num_points);
//
// for (n = 0; n < num_out_vectors; n++)
// {
// volk_gnsssdr_free(result_aux[n]);
// }
// volk_gnsssdr_free(result_aux);
//}
//#endif
#ifdef LV_HAVE_AVX
static inline void volk_gnsssdr_32f_fast_resamplerxnpuppet_32f_a_avx(float* result, const float* local_code, unsigned int num_points)
{
int code_length_chips = 2046;
float code_phase_step_chips = ((float)(code_length_chips) + 0.1) / ((float)num_points);
int num_out_vectors = 3;
float rem_code_phase_chips = -0.8234;
float code_phase_rate_step_chips = 1.0 / powf(2.0, 33.0);
unsigned int n;
float shifts_chips[3] = {-0.1, 0.0, 0.1};
float** result_aux = (float**)volk_gnsssdr_malloc(sizeof(float*) * num_out_vectors, volk_gnsssdr_get_alignment());
for (n = 0; n < num_out_vectors; n++)
{
result_aux[n] = (float*)volk_gnsssdr_malloc(sizeof(float) * num_points, volk_gnsssdr_get_alignment());
}
volk_gnsssdr_32f_xn_fast_resampler_32f_xn_a_avx(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, code_phase_rate_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
memcpy((float*)result, (float*)result_aux[0], sizeof(float) * num_points);
for (n = 0; n < num_out_vectors; n++)
{
volk_gnsssdr_free(result_aux[n]);
}
volk_gnsssdr_free(result_aux);
}
#endif
#ifdef LV_HAVE_AVX
static inline void volk_gnsssdr_32f_fast_resamplerxnpuppet_32f_u_avx(float* result, const float* local_code, unsigned int num_points)
{
int code_length_chips = 2046;
float code_phase_step_chips = ((float)(code_length_chips) + 0.1) / ((float)num_points);
int num_out_vectors = 3;
float rem_code_phase_chips = -0.8234;
float code_phase_rate_step_chips = 1.0 / powf(2.0, 33.0);
unsigned int n;
float shifts_chips[3] = {-0.1, 0.0, 0.1};
float** result_aux = (float**)volk_gnsssdr_malloc(sizeof(float*) * num_out_vectors, volk_gnsssdr_get_alignment());
for (n = 0; n < num_out_vectors; n++)
{
result_aux[n] = (float*)volk_gnsssdr_malloc(sizeof(float) * num_points, volk_gnsssdr_get_alignment());
}
volk_gnsssdr_32f_xn_fast_resampler_32f_xn_u_avx(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, code_phase_rate_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
memcpy((float*)result, (float*)result_aux[0], sizeof(float) * num_points);
for (n = 0; n < num_out_vectors; n++)
{
volk_gnsssdr_free(result_aux[n]);
}
volk_gnsssdr_free(result_aux);
}
#endif
//
//#ifdef LV_HAVE_NEONV7
//static inline void volk_gnsssdr_32f_resamplerxnpuppet_32f_neon(float* result, const float* local_code, unsigned int num_points)

332
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_32f_xn_fast_resampler_32f_xn.h
View File

@ -440,162 +440,182 @@ static inline void volk_gnsssdr_32f_xn_fast_resampler_32f_xn_u_sse4_1(float** re
}
#endif
//
//
//#ifdef LV_HAVE_AVX
//#include <immintrin.h>
//static inline void volk_gnsssdr_32f_xn_fast_resampler_32f_xn_a_avx(float** result, const float* local_code, float rem_code_phase_chips, float code_phase_step_chips, float* shifts_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_points)
//{
// float** _result = result;
// const unsigned int avx_iters = num_points / 8;
// int current_correlator_tap;
// unsigned int n;
// unsigned int k;
// const __m256 eights = _mm256_set1_ps(8.0f);
// const __m256 rem_code_phase_chips_reg = _mm256_set1_ps(rem_code_phase_chips);
// const __m256 code_phase_step_chips_reg = _mm256_set1_ps(code_phase_step_chips);
//
// __VOLK_ATTR_ALIGNED(32)
// int local_code_chip_index[8];
// int local_code_chip_index_;
//
// const __m256 zeros = _mm256_setzero_ps();
// const __m256 code_length_chips_reg_f = _mm256_set1_ps((float)code_length_chips);
// const __m256 n0 = _mm256_set_ps(7.0f, 6.0f, 5.0f, 4.0f, 3.0f, 2.0f, 1.0f, 0.0f);
//
// __m256i local_code_chip_index_reg, i;
// __m256 aux, aux2, aux3, shifts_chips_reg, c, cTrunc, base, negatives, indexn;
//
// for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
// {
// shifts_chips_reg = _mm256_set1_ps((float)shifts_chips[current_correlator_tap]);
// aux2 = _mm256_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
// indexn = n0;
// for (n = 0; n < avx_iters; n++)
// {
// __VOLK_GNSSSDR_PREFETCH_LOCALITY(&_result[current_correlator_tap][8 * n + 7], 1, 0);
// __VOLK_GNSSSDR_PREFETCH_LOCALITY(&local_code_chip_index[8], 1, 3);
// aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
// aux = _mm256_add_ps(aux, aux2);
// // floor
// aux = _mm256_floor_ps(aux);
//
// // fmod
// c = _mm256_div_ps(aux, code_length_chips_reg_f);
// i = _mm256_cvttps_epi32(c);
// cTrunc = _mm256_cvtepi32_ps(i);
// base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
// local_code_chip_index_reg = _mm256_cvttps_epi32(_mm256_sub_ps(aux, base));
//
// // no negatives
// c = _mm256_cvtepi32_ps(local_code_chip_index_reg);
// negatives = _mm256_cmp_ps(c, zeros, 0x01);
// aux3 = _mm256_and_ps(code_length_chips_reg_f, negatives);
// aux = _mm256_add_ps(c, aux3);
// local_code_chip_index_reg = _mm256_cvttps_epi32(aux);
//
// _mm256_store_si256((__m256i*)local_code_chip_index, local_code_chip_index_reg);
// for (k = 0; k < 8; ++k)
// {
// _result[current_correlator_tap][n * 8 + k] = local_code[local_code_chip_index[k]];
// }
// indexn = _mm256_add_ps(indexn, eights);
// }
// }
// _mm256_zeroupper();
// for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
// {
// for (n = avx_iters * 8; n < num_points; n++)
// {
// // resample code for current tap
// local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
// //Take into account that in multitap correlators, the shifts can be negative!
// if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1);
// local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
// _result[current_correlator_tap][n] = local_code[local_code_chip_index_];
// }
// }
//}
//
//#endif
//
//
//#ifdef LV_HAVE_AVX
//#include <immintrin.h>
//static inline void volk_gnsssdr_32f_xn_fast_resampler_32f_xn_u_avx(float** result, const float* local_code, float rem_code_phase_chips, float code_phase_step_chips, float* shifts_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_points)
//{
// float** _result = result;
// const unsigned int avx_iters = num_points / 8;
// int current_correlator_tap;
// unsigned int n;
// unsigned int k;
// const __m256 eights = _mm256_set1_ps(8.0f);
// const __m256 rem_code_phase_chips_reg = _mm256_set1_ps(rem_code_phase_chips);
// const __m256 code_phase_step_chips_reg = _mm256_set1_ps(code_phase_step_chips);
//
// __VOLK_ATTR_ALIGNED(32)
// int local_code_chip_index[8];
// int local_code_chip_index_;
//
// const __m256 zeros = _mm256_setzero_ps();
// const __m256 code_length_chips_reg_f = _mm256_set1_ps((float)code_length_chips);
// const __m256 n0 = _mm256_set_ps(7.0f, 6.0f, 5.0f, 4.0f, 3.0f, 2.0f, 1.0f, 0.0f);
//
// __m256i local_code_chip_index_reg, i;
// __m256 aux, aux2, aux3, shifts_chips_reg, c, cTrunc, base, negatives, indexn;
//
// for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
// {
// shifts_chips_reg = _mm256_set1_ps((float)shifts_chips[current_correlator_tap]);
// aux2 = _mm256_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
// indexn = n0;
// for (n = 0; n < avx_iters; n++)
// {
// __VOLK_GNSSSDR_PREFETCH_LOCALITY(&_result[current_correlator_tap][8 * n + 7], 1, 0);
// __VOLK_GNSSSDR_PREFETCH_LOCALITY(&local_code_chip_index[8], 1, 3);
// aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
// aux = _mm256_add_ps(aux, aux2);
// // floor
// aux = _mm256_floor_ps(aux);
//
// // fmod
// c = _mm256_div_ps(aux, code_length_chips_reg_f);
// i = _mm256_cvttps_epi32(c);
// cTrunc = _mm256_cvtepi32_ps(i);
// base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
// local_code_chip_index_reg = _mm256_cvttps_epi32(_mm256_sub_ps(aux, base));
//
// // no negatives
// c = _mm256_cvtepi32_ps(local_code_chip_index_reg);
// negatives = _mm256_cmp_ps(c, zeros, 0x01);
// aux3 = _mm256_and_ps(code_length_chips_reg_f, negatives);
// aux = _mm256_add_ps(c, aux3);
// local_code_chip_index_reg = _mm256_cvttps_epi32(aux);
//
// _mm256_store_si256((__m256i*)local_code_chip_index, local_code_chip_index_reg);
// for (k = 0; k < 8; ++k)
// {
// _result[current_correlator_tap][n * 8 + k] = local_code[local_code_chip_index[k]];
// }
// indexn = _mm256_add_ps(indexn, eights);
// }
// }
// _mm256_zeroupper();
// for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
// {
// for (n = avx_iters * 8; n < num_points; n++)
// {
// // resample code for current tap
// local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
// //Take into account that in multitap correlators, the shifts can be negative!
// if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1);
// local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
// _result[current_correlator_tap][n] = local_code[local_code_chip_index_];
// }
// }
//}
//
//#endif
#ifdef LV_HAVE_AVX
#include <immintrin.h>
static inline void volk_gnsssdr_32f_xn_fast_resampler_32f_xn_a_avx(float** result, const float* local_code, float rem_code_phase_chips, float code_phase_step_chips, float code_phase_rate_step_chips, float* shifts_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_points)
{
float** _result = result;
const unsigned int avx_iters = num_points / 8;
int current_correlator_tap;
unsigned int n;
unsigned int k;
const __m256 eights = _mm256_set1_ps(8.0f);
const __m256 ones = _mm256_set1_ps(1.0f);
const __m256 rem_code_phase_chips_reg = _mm256_set1_ps(rem_code_phase_chips);
const __m256 code_phase_step_chips_reg = _mm256_set1_ps(code_phase_step_chips);
const __m256 code_phase_rate_step_chips_reg = _mm256_set1_ps(code_phase_rate_step_chips);
__VOLK_ATTR_ALIGNED(32)
int local_code_chip_index[8];
int local_code_chip_index_;
const __m256 zeros = _mm256_setzero_ps();
const __m256 code_length_chips_reg_f = _mm256_set1_ps((float)code_length_chips);
const __m256 n0 = _mm256_set_ps(7.0f, 6.0f, 5.0f, 4.0f, 3.0f, 2.0f, 1.0f, 0.0f);
__m256i local_code_chip_index_reg, i;
__m256 aux, aux2, aux3, shifts_chips_reg, c, cTrunc, base, negatives, indexn, indexnn;
shifts_chips_reg = _mm256_set1_ps((float)shifts_chips[0]);
aux2 = _mm256_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
indexn = n0;
for (n = 0; n < avx_iters; n++)
{
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&_result[0][8 * n + 7], 1, 0);
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&local_code_chip_index[8], 1, 3);
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
indexnn = _mm256_mul_ps(indexn, indexn);
aux3 = _mm256_mul_ps(code_phase_rate_step_chips_reg, indexnn);
aux = _mm256_add_ps(aux, aux3);
aux = _mm256_add_ps(aux, aux2);
// floor
aux = _mm256_floor_ps(aux);
// Correct negative shift
c = _mm256_div_ps(aux, code_length_chips_reg_f);
aux3 = _mm256_add_ps(c, ones);
i = _mm256_cvttps_epi32(aux3);
cTrunc = _mm256_cvtepi32_ps(i);
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
local_code_chip_index_reg = _mm256_cvttps_epi32(_mm256_sub_ps(aux, base));
c = _mm256_cvtepi32_ps(local_code_chip_index_reg);
negatives = _mm256_cmp_ps(c, zeros, 0x01);
aux3 = _mm256_and_ps(code_length_chips_reg_f, negatives);
aux = _mm256_add_ps(c, aux3);
local_code_chip_index_reg = _mm256_cvttps_epi32(aux);
_mm256_store_si256((__m256i*)local_code_chip_index, local_code_chip_index_reg);
for (k = 0; k < 8; ++k)
{
_result[0][n * 8 + k] = local_code[local_code_chip_index[k]];
}
indexn = _mm256_add_ps(indexn, eights);
}
_mm256_zeroupper();
for (n = avx_iters * 8; n < num_points; n++)
{
// resample code for first tap
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + code_phase_rate_step_chips * (float)(n * n) + shifts_chips[0] - rem_code_phase_chips);
// Take into account that in multitap correlators, the shifts can be negative!
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1);
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
_result[0][n] = local_code[local_code_chip_index_];
}
// adjacent correlators
unsigned int shift_samples = 0;
for (current_correlator_tap = 1; current_correlator_tap < num_out_vectors; current_correlator_tap++)
{
shift_samples += (int)round((shifts_chips[current_correlator_tap] - shifts_chips[current_correlator_tap - 1]) / code_phase_step_chips);
memcpy(&_result[current_correlator_tap][0], &_result[0][shift_samples], (num_points - shift_samples) * sizeof(float));
memcpy(&_result[current_correlator_tap][num_points - shift_samples], &_result[0][0], shift_samples * sizeof(float));
}
}
#endif
#ifdef LV_HAVE_AVX
#include <immintrin.h>
static inline void volk_gnsssdr_32f_xn_fast_resampler_32f_xn_u_avx(float** result, const float* local_code, float rem_code_phase_chips, float code_phase_step_chips, float code_phase_rate_step_chips, float* shifts_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_points)
{
float** _result = result;
const unsigned int avx_iters = num_points / 8;
int current_correlator_tap;
unsigned int n;
unsigned int k;
const __m256 eights = _mm256_set1_ps(8.0f);
const __m256 ones = _mm256_set1_ps(1.0f);
const __m256 rem_code_phase_chips_reg = _mm256_set1_ps(rem_code_phase_chips);
const __m256 code_phase_step_chips_reg = _mm256_set1_ps(code_phase_step_chips);
const __m256 code_phase_rate_step_chips_reg = _mm256_set1_ps(code_phase_rate_step_chips);
__VOLK_ATTR_ALIGNED(32)
int local_code_chip_index[8];
int local_code_chip_index_;
const __m256 zeros = _mm256_setzero_ps();
const __m256 code_length_chips_reg_f = _mm256_set1_ps((float)code_length_chips);
const __m256 n0 = _mm256_set_ps(7.0f, 6.0f, 5.0f, 4.0f, 3.0f, 2.0f, 1.0f, 0.0f);
__m256i local_code_chip_index_reg, i;
__m256 aux, aux2, aux3, shifts_chips_reg, c, cTrunc, base, negatives, indexn, indexnn;
shifts_chips_reg = _mm256_set1_ps((float)shifts_chips[0]);
aux2 = _mm256_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
indexn = n0;
for (n = 0; n < avx_iters; n++)
{
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&_result[0][8 * n + 7], 1, 0);
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&local_code_chip_index[8], 1, 3);
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
indexnn = _mm256_mul_ps(indexn, indexn);
aux3 = _mm256_mul_ps(code_phase_rate_step_chips_reg, indexnn);
aux = _mm256_add_ps(aux, aux3);
aux = _mm256_add_ps(aux, aux2);
// floor
aux = _mm256_floor_ps(aux);
// Correct negative shift
c = _mm256_div_ps(aux, code_length_chips_reg_f);
aux3 = _mm256_add_ps(c, ones);
i = _mm256_cvttps_epi32(aux3);
cTrunc = _mm256_cvtepi32_ps(i);
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
local_code_chip_index_reg = _mm256_cvttps_epi32(_mm256_sub_ps(aux, base));
c = _mm256_cvtepi32_ps(local_code_chip_index_reg);
negatives = _mm256_cmp_ps(c, zeros, 0x01);
aux3 = _mm256_and_ps(code_length_chips_reg_f, negatives);
aux = _mm256_add_ps(c, aux3);
local_code_chip_index_reg = _mm256_cvttps_epi32(aux);
_mm256_store_si256((__m256i*)local_code_chip_index, local_code_chip_index_reg);
for (k = 0; k < 8; ++k)
{
_result[0][n * 8 + k] = local_code[local_code_chip_index[k]];
}
indexn = _mm256_add_ps(indexn, eights);
}
_mm256_zeroupper();
for (n = avx_iters * 8; n < num_points; n++)
{
// resample code for first tap
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + code_phase_rate_step_chips * (float)(n * n) + shifts_chips[0] - rem_code_phase_chips);
// Take into account that in multitap correlators, the shifts can be negative!
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1);
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
_result[0][n] = local_code[local_code_chip_index_];
}
// adjacent correlators
unsigned int shift_samples = 0;
for (current_correlator_tap = 1; current_correlator_tap < num_out_vectors; current_correlator_tap++)
{
shift_samples += (int)round((shifts_chips[current_correlator_tap] - shifts_chips[current_correlator_tap - 1]) / code_phase_step_chips);
memcpy(&_result[current_correlator_tap][0], &_result[0][shift_samples], (num_points - shift_samples) * sizeof(float));
memcpy(&_result[current_correlator_tap][num_points - shift_samples], &_result[0][0], shift_samples * sizeof(float));
}
}
#endif
//
//
//#ifdef LV_HAVE_NEONV7