mirror of
https://github.com/gnss-sdr/gnss-sdr
synced 2025-10-24 20:17:39 +00:00
Add Doppler rate in fast_resampler kernel. Still not used
This commit is contained in:
Carles Fernandez
@@ -49,7 +49,8 @@ static inline void volk_gnsssdr_32f_fast_resamplerxnpuppet_32f_generic(float* re
|
||||
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;
|
||||
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};
|
||||
|
||||
@@ -59,7 +60,7 @@ static inline void volk_gnsssdr_32f_fast_resamplerxnpuppet_32f_generic(float* re
|
||||
result_aux[n] = (float*)volk_gnsssdr_malloc(sizeof(float) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_32f_xn_fast_resampler_32f_xn_generic(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
volk_gnsssdr_32f_xn_fast_resampler_32f_xn_generic(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);
|
||||
|
||||
@@ -73,63 +74,65 @@ static inline void volk_gnsssdr_32f_fast_resamplerxnpuppet_32f_generic(float* re
|
||||
|
||||
#endif /* LV_HAVE_GENERIC */
|
||||
|
||||
//#ifdef LV_HAVE_SSE3
|
||||
//static inline void volk_gnsssdr_32f_resamplerxnpuppet_32f_a_sse3(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_sse3(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_SSE3
|
||||
//static inline void volk_gnsssdr_32f_resamplerxnpuppet_32f_u_sse3(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_sse3(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_SSE3
|
||||
static inline void volk_gnsssdr_32f_fast_resamplerxnpuppet_32f_a_sse3(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_sse3(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_SSE3
|
||||
static inline void volk_gnsssdr_32f_fast_resamplerxnpuppet_32f_u_sse3(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_sse3(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_SSE4_1
|
||||
|
||||
@@ -46,20 +46,21 @@
|
||||
*
|
||||
* <b>Dispatcher Prototype</b>
|
||||
* \code
|
||||
* void volk_gnsssdr_32f_xn_fast_resampler_32f_xn(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)
|
||||
* void volk_gnsssdr_32f_xn_fast_resampler_32f_xn(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)
|
||||
* \endcode
|
||||
*
|
||||
* \b Inputs
|
||||
* \li local_code: Vector to be resampled.
|
||||
* \li rem_code_phase_chips: Remnant code phase [chips].
|
||||
* \li code_phase_step_chips: Phase increment per sample [chips/sample].
|
||||
* \li shifts_chips: Vector of floats that defines the spacing (in chips) between the replicas of \p local_code
|
||||
* \li code_length_chips: Code length in chips.
|
||||
* \li num_out_vectors Number of output vectors.
|
||||
* \li num_points: The number of data values to be in the resampled vector.
|
||||
* \li local_code: Vector to be resampled.
|
||||
* \li rem_code_phase_chips: Remnant code phase [chips].
|
||||
* \li code_phase_step_chips: Phase increment per sample [chips/sample].
|
||||
* \li code_phase_rate_step_chips: Phase rate increment per sample [chips/sample^2].
|
||||
* \li shifts_chips: Vector of floats that defines the spacing (in chips) between the replicas of \p local_code
|
||||
* \li code_length_chips: Code length in chips.
|
||||
* \li num_out_vectors Number of output vectors.
|
||||
* \li num_points: The number of data values to be in the resampled vector.
|
||||
*
|
||||
* \b Outputs
|
||||
* \li result: Pointer to a vector of pointers where the results will be stored.
|
||||
* \li result: Pointer to a vector of pointers where the results will be stored.
|
||||
*
|
||||
*/
|
||||
|
||||
@@ -77,7 +78,7 @@
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
|
||||
static inline void volk_gnsssdr_32f_xn_fast_resampler_32f_xn_generic(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)
|
||||
static inline void volk_gnsssdr_32f_xn_fast_resampler_32f_xn_generic(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)
|
||||
{
|
||||
int local_code_chip_index;
|
||||
int current_correlator_tap;
|
||||
@@ -85,9 +86,9 @@ static inline void volk_gnsssdr_32f_xn_fast_resampler_32f_xn_generic(float** res
|
||||
//first correlator
|
||||
for (n = 0; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index = (int)floor(code_phase_step_chips * (float)n + shifts_chips[0] - rem_code_phase_chips);
|
||||
//Take into account that in multitap correlators, the shifts can be negative!
|
||||
// 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];
|
||||
@@ -106,145 +107,175 @@ static inline void volk_gnsssdr_32f_xn_fast_resampler_32f_xn_generic(float** res
|
||||
#endif /*LV_HAVE_GENERIC*/
|
||||
|
||||
|
||||
//#ifdef LV_HAVE_SSE3
|
||||
//#include <pmmintrin.h>
|
||||
//static inline void volk_gnsssdr_32f_xn_fast_resampler_32f_xn_a_sse3(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 quarterPoints = num_points / 4;
|
||||
// int current_correlator_tap;
|
||||
// unsigned int n;
|
||||
// unsigned int k;
|
||||
// const __m128 ones = _mm_set1_ps(1.0f);
|
||||
// const __m128 fours = _mm_set1_ps(4.0f);
|
||||
// const __m128 rem_code_phase_chips_reg = _mm_set_ps1(rem_code_phase_chips);
|
||||
// const __m128 code_phase_step_chips_reg = _mm_set_ps1(code_phase_step_chips);
|
||||
//
|
||||
// __VOLK_ATTR_ALIGNED(16)
|
||||
// int local_code_chip_index[4];
|
||||
// int local_code_chip_index_;
|
||||
//
|
||||
// const __m128i zeros = _mm_setzero_si128();
|
||||
// const __m128 code_length_chips_reg_f = _mm_set_ps1((float)code_length_chips);
|
||||
// const __m128i code_length_chips_reg_i = _mm_set1_epi32((int)code_length_chips);
|
||||
// __m128i local_code_chip_index_reg, aux_i, negatives, i;
|
||||
// __m128 aux, aux2, shifts_chips_reg, fi, igx, j, c, cTrunc, base;
|
||||
//
|
||||
// for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
// {
|
||||
// shifts_chips_reg = _mm_set_ps1((float)shifts_chips[current_correlator_tap]);
|
||||
// aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
// __m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
|
||||
// for (n = 0; n < quarterPoints; n++)
|
||||
// {
|
||||
// aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
// aux = _mm_add_ps(aux, aux2);
|
||||
// // floor
|
||||
// i = _mm_cvttps_epi32(aux);
|
||||
// fi = _mm_cvtepi32_ps(i);
|
||||
// igx = _mm_cmpgt_ps(fi, aux);
|
||||
// j = _mm_and_ps(igx, ones);
|
||||
// aux = _mm_sub_ps(fi, j);
|
||||
// // fmod
|
||||
// c = _mm_div_ps(aux, code_length_chips_reg_f);
|
||||
// i = _mm_cvttps_epi32(c);
|
||||
// cTrunc = _mm_cvtepi32_ps(i);
|
||||
// base = _mm_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||
// local_code_chip_index_reg = _mm_cvtps_epi32(_mm_sub_ps(aux, base));
|
||||
//
|
||||
// negatives = _mm_cmplt_epi32(local_code_chip_index_reg, zeros);
|
||||
// aux_i = _mm_and_si128(code_length_chips_reg_i, negatives);
|
||||
// local_code_chip_index_reg = _mm_add_epi32(local_code_chip_index_reg, aux_i);
|
||||
// _mm_store_si128((__m128i*)local_code_chip_index, local_code_chip_index_reg);
|
||||
// for (k = 0; k < 4; ++k)
|
||||
// {
|
||||
// _result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
// }
|
||||
// indexn = _mm_add_ps(indexn, fours);
|
||||
// }
|
||||
// for (n = quarterPoints * 4; 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_SSE3
|
||||
//#include <pmmintrin.h>
|
||||
//static inline void volk_gnsssdr_32f_xn_fast_resampler_32f_xn_u_sse3(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 quarterPoints = num_points / 4;
|
||||
// int current_correlator_tap;
|
||||
// unsigned int n;
|
||||
// unsigned int k;
|
||||
// const __m128 ones = _mm_set1_ps(1.0f);
|
||||
// const __m128 fours = _mm_set1_ps(4.0f);
|
||||
// const __m128 rem_code_phase_chips_reg = _mm_set_ps1(rem_code_phase_chips);
|
||||
// const __m128 code_phase_step_chips_reg = _mm_set_ps1(code_phase_step_chips);
|
||||
//
|
||||
// __VOLK_ATTR_ALIGNED(16)
|
||||
// int local_code_chip_index[4];
|
||||
// int local_code_chip_index_;
|
||||
//
|
||||
// const __m128i zeros = _mm_setzero_si128();
|
||||
// const __m128 code_length_chips_reg_f = _mm_set_ps1((float)code_length_chips);
|
||||
// const __m128i code_length_chips_reg_i = _mm_set1_epi32((int)code_length_chips);
|
||||
// __m128i local_code_chip_index_reg, aux_i, negatives, i;
|
||||
// __m128 aux, aux2, shifts_chips_reg, fi, igx, j, c, cTrunc, base;
|
||||
//
|
||||
// for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
// {
|
||||
// shifts_chips_reg = _mm_set_ps1((float)shifts_chips[current_correlator_tap]);
|
||||
// aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
// __m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
|
||||
// for (n = 0; n < quarterPoints; n++)
|
||||
// {
|
||||
// aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
// aux = _mm_add_ps(aux, aux2);
|
||||
// // floor
|
||||
// i = _mm_cvttps_epi32(aux);
|
||||
// fi = _mm_cvtepi32_ps(i);
|
||||
// igx = _mm_cmpgt_ps(fi, aux);
|
||||
// j = _mm_and_ps(igx, ones);
|
||||
// aux = _mm_sub_ps(fi, j);
|
||||
// // fmod
|
||||
// c = _mm_div_ps(aux, code_length_chips_reg_f);
|
||||
// i = _mm_cvttps_epi32(c);
|
||||
// cTrunc = _mm_cvtepi32_ps(i);
|
||||
// base = _mm_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||
// local_code_chip_index_reg = _mm_cvtps_epi32(_mm_sub_ps(aux, base));
|
||||
//
|
||||
// negatives = _mm_cmplt_epi32(local_code_chip_index_reg, zeros);
|
||||
// aux_i = _mm_and_si128(code_length_chips_reg_i, negatives);
|
||||
// local_code_chip_index_reg = _mm_add_epi32(local_code_chip_index_reg, aux_i);
|
||||
// _mm_store_si128((__m128i*)local_code_chip_index, local_code_chip_index_reg);
|
||||
// for (k = 0; k < 4; ++k)
|
||||
// {
|
||||
// _result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
// }
|
||||
// indexn = _mm_add_ps(indexn, fours);
|
||||
// }
|
||||
// for (n = quarterPoints * 4; 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_SSE3
|
||||
#include <pmmintrin.h>
|
||||
static inline void volk_gnsssdr_32f_xn_fast_resampler_32f_xn_a_sse3(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 quarterPoints = num_points / 4;
|
||||
// int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
unsigned int current_correlator_tap;
|
||||
const __m128 ones = _mm_set1_ps(1.0f);
|
||||
const __m128 fours = _mm_set1_ps(4.0f);
|
||||
const __m128 rem_code_phase_chips_reg = _mm_set_ps1(rem_code_phase_chips);
|
||||
const __m128 code_phase_step_chips_reg = _mm_set_ps1(code_phase_step_chips);
|
||||
const __m128 code_phase_rate_step_chips_reg = _mm_set_ps1(code_phase_rate_step_chips);
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16)
|
||||
int local_code_chip_index[4];
|
||||
int local_code_chip_index_;
|
||||
const __m128i zeros = _mm_setzero_si128();
|
||||
const __m128 code_length_chips_reg_f = _mm_set_ps1((float)code_length_chips);
|
||||
const __m128i code_length_chips_reg_i = _mm_set1_epi32((int)code_length_chips);
|
||||
__m128i local_code_chip_index_reg, aux_i, negatives;
|
||||
__m128 aux, aux2, aux3, indexnn, shifts_chips_reg, i, fi, igx, j, c, cTrunc, base;
|
||||
__m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
|
||||
|
||||
shifts_chips_reg = _mm_set_ps1((float)shifts_chips[0]);
|
||||
aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
|
||||
for (n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
indexnn = _mm_mul_ps(indexn, indexn);
|
||||
aux3 = _mm_mul_ps(code_phase_rate_step_chips_reg, indexnn);
|
||||
aux = _mm_add_ps(aux, aux3);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
i = _mm_cvttps_epi32(aux);
|
||||
fi = _mm_cvtepi32_ps(i);
|
||||
igx = _mm_cmpgt_ps(fi, aux);
|
||||
j = _mm_and_ps(igx, ones);
|
||||
aux = _mm_sub_ps(fi, j);
|
||||
|
||||
// Correct negative shift
|
||||
c = _mm_div_ps(aux, code_length_chips_reg_f);
|
||||
aux3 = _mm_add_ps(c, ones);
|
||||
i = _mm_cvttps_epi32(aux3);
|
||||
cTrunc = _mm_cvtepi32_ps(i);
|
||||
base = _mm_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||
local_code_chip_index_reg = _mm_cvtps_epi32(_mm_sub_ps(aux, base));
|
||||
negatives = _mm_cmplt_epi32(local_code_chip_index_reg, zeros);
|
||||
aux_i = _mm_and_si128(code_length_chips_reg_i, negatives);
|
||||
local_code_chip_index_reg = _mm_add_epi32(local_code_chip_index_reg, aux_i);
|
||||
|
||||
_mm_store_si128((__m128i*)local_code_chip_index, local_code_chip_index_reg);
|
||||
|
||||
for (k = 0; k < 4; ++k)
|
||||
{
|
||||
_result[0][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm_add_ps(indexn, fours);
|
||||
}
|
||||
|
||||
for (n = quarterPoints * 4; 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_SSE3
|
||||
#include <pmmintrin.h>
|
||||
static inline void volk_gnsssdr_32f_xn_fast_resampler_32f_xn_u_sse3(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 quarterPoints = num_points / 4;
|
||||
// int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
unsigned int current_correlator_tap;
|
||||
const __m128 ones = _mm_set1_ps(1.0f);
|
||||
const __m128 fours = _mm_set1_ps(4.0f);
|
||||
const __m128 rem_code_phase_chips_reg = _mm_set_ps1(rem_code_phase_chips);
|
||||
const __m128 code_phase_step_chips_reg = _mm_set_ps1(code_phase_step_chips);
|
||||
const __m128 code_phase_rate_step_chips_reg = _mm_set_ps1(code_phase_rate_step_chips);
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16)
|
||||
int local_code_chip_index[4];
|
||||
int local_code_chip_index_;
|
||||
const __m128i zeros = _mm_setzero_si128();
|
||||
const __m128 code_length_chips_reg_f = _mm_set_ps1((float)code_length_chips);
|
||||
const __m128i code_length_chips_reg_i = _mm_set1_epi32((int)code_length_chips);
|
||||
__m128i local_code_chip_index_reg, aux_i, negatives;
|
||||
__m128 aux, aux2, aux3, indexnn, shifts_chips_reg, i, fi, igx, j, c, cTrunc, base;
|
||||
__m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
|
||||
|
||||
shifts_chips_reg = _mm_set_ps1((float)shifts_chips[0]);
|
||||
aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
|
||||
for (n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
indexnn = _mm_mul_ps(indexn, indexn);
|
||||
aux3 = _mm_mul_ps(code_phase_rate_step_chips_reg, indexnn);
|
||||
aux = _mm_add_ps(aux, aux3);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
i = _mm_cvttps_epi32(aux);
|
||||
fi = _mm_cvtepi32_ps(i);
|
||||
igx = _mm_cmpgt_ps(fi, aux);
|
||||
j = _mm_and_ps(igx, ones);
|
||||
aux = _mm_sub_ps(fi, j);
|
||||
|
||||
// Correct negative shift
|
||||
c = _mm_div_ps(aux, code_length_chips_reg_f);
|
||||
aux3 = _mm_add_ps(c, ones);
|
||||
i = _mm_cvttps_epi32(aux3);
|
||||
cTrunc = _mm_cvtepi32_ps(i);
|
||||
base = _mm_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||
local_code_chip_index_reg = _mm_cvtps_epi32(_mm_sub_ps(aux, base));
|
||||
negatives = _mm_cmplt_epi32(local_code_chip_index_reg, zeros);
|
||||
aux_i = _mm_and_si128(code_length_chips_reg_i, negatives);
|
||||
local_code_chip_index_reg = _mm_add_epi32(local_code_chip_index_reg, aux_i);
|
||||
|
||||
_mm_store_si128((__m128i*)local_code_chip_index, local_code_chip_index_reg);
|
||||
|
||||
for (k = 0; k < 4; ++k)
|
||||
{
|
||||
_result[0][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm_add_ps(indexn, fours);
|
||||
}
|
||||
|
||||
for (n = quarterPoints * 4; 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_SSE4_1
|
||||
|
||||
@@ -98,7 +98,7 @@ bool cpu_multicorrelator_real_codes::set_input_output_vectors(std::complex<float
|
||||
}
|
||||
|
||||
|
||||
void cpu_multicorrelator_real_codes::update_local_code(int correlator_length_samples, float rem_code_phase_chips, float code_phase_step_chips)
|
||||
void cpu_multicorrelator_real_codes::update_local_code(int correlator_length_samples, float rem_code_phase_chips, float code_phase_step_chips, float code_phase_rate_step_chips)
|
||||
{
|
||||
if (d_use_fast_resampler)
|
||||
{
|
||||
@@ -106,6 +106,7 @@ void cpu_multicorrelator_real_codes::update_local_code(int correlator_length_sam
|
||||
d_local_code_in,
|
||||
rem_code_phase_chips,
|
||||
code_phase_step_chips,
|
||||
code_phase_rate_step_chips,
|
||||
d_shifts_chips,
|
||||
d_code_length_chips,
|
||||
d_n_correlators,
|
||||
|
||||
@@ -51,7 +51,7 @@ public:
|
||||
bool init(int max_signal_length_samples, int n_correlators);
|
||||
bool set_local_code_and_taps(int code_length_chips, const float *local_code_in, float *shifts_chips);
|
||||
bool set_input_output_vectors(std::complex<float> *corr_out, const std::complex<float> *sig_in);
|
||||
void update_local_code(int correlator_length_samples, float rem_code_phase_chips, float code_phase_step_chips);
|
||||
void update_local_code(int correlator_length_samples, float rem_code_phase_chips, float code_phase_step_chips, float code_phase_rate_step_chips = 0.0);
|
||||
bool Carrier_wipeoff_multicorrelator_resampler(float rem_carrier_phase_in_rad, float phase_step_rad, float rem_code_phase_chips, float code_phase_step_chips, int signal_length_samples);
|
||||
bool free();
|
||||
|
||||
|
||||
Reference in New Issue
Block a user