mirror of
https://github.com/gnss-sdr/gnss-sdr
synced 2024-06-24 22:13:15 +00:00
More consistent naming
now volk_gnsssdr_16ic_xn_resampler_16ic implements the same resampler than volk_gnsssdr_32fc_xn_resampler_32fc. The old one, which is faster in SSE implementations at the expense of some constraints on the inputs (to be documented), is now named volk_gnsssdr_16ic_xn_resampler_fast_16ic
This commit is contained in:
Carles Fernandez
parent
4c4f6c1def
commit
25e5c744d7
|
|
@ -1,5 +1,5 @@
|
|||
/*!
|
||||
* \file volk_gnsssdr_16ic_resampler_16ic.h
|
||||
* \file volk_gnsssdr_16ic_resampler_fast_16ic.h
|
||||
* \brief VOLK_GNSSSDR kernel: resamples a 16 bits complex vector.
|
||||
* \authors <ul>
|
||||
* <li> Javier Arribas, 2015. jarribas(at)cttc.es
|
||||
|
|
@ -34,7 +34,7 @@
|
|||
*/
|
||||
|
||||
/*!
|
||||
* \page volk_gnsssdr_16ic_resampler_16ic
|
||||
* \page volk_gnsssdr_16ic_resampler_fast_16ic
|
||||
*
|
||||
* \b Overview
|
||||
*
|
||||
|
|
@ -42,7 +42,7 @@
|
|||
*
|
||||
* <b>Dispatcher Prototype</b>
|
||||
* \code
|
||||
* void volk_gnsssdr_16ic_resampler_16ic(lv_16sc_t* result, const lv_16sc_t* local_code, float rem_code_phase_chips, float code_phase_step_chips, int code_length_chips, unsigned int num_output_samples)
|
||||
* void volk_gnsssdr_16ic_resampler_fast_16ic(lv_16sc_t* result, const lv_16sc_t* local_code, float rem_code_phase_chips, float code_phase_step_chips, int code_length_chips, unsigned int num_output_samples)
|
||||
* \endcode
|
||||
*
|
||||
* \b Inputs
|
||||
|
|
@ -57,8 +57,8 @@
|
|||
*
|
||||
*/
|
||||
|
||||
#ifndef INCLUDED_volk_gnsssdr_16ic_resampler_16ic_H
|
||||
#define INCLUDED_volk_gnsssdr_16ic_resampler_16ic_H
|
||||
#ifndef INCLUDED_volk_gnsssdr_16ic_resampler_fast_16ic_H
|
||||
#define INCLUDED_volk_gnsssdr_16ic_resampler_fast_16ic_H
|
||||
|
||||
#include <math.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_common.h>
|
||||
|
|
@ -72,7 +72,7 @@
|
|||
// return (r > 0.0) ? (r + 0.5) : (r - 0.5);
|
||||
//}
|
||||
|
||||
static inline void volk_gnsssdr_16ic_resampler_16ic_generic(lv_16sc_t* result, const lv_16sc_t* local_code, float rem_code_phase_chips, float code_phase_step_chips, int code_length_chips, unsigned int num_output_samples)
|
||||
static inline void volk_gnsssdr_16ic_resampler_fast_16ic_generic(lv_16sc_t* result, const lv_16sc_t* local_code, float rem_code_phase_chips, float code_phase_step_chips, int code_length_chips, unsigned int num_output_samples)
|
||||
{
|
||||
int local_code_chip_index;
|
||||
//fesetround(FE_TONEAREST);
|
||||
|
|
@ -92,7 +92,7 @@ static inline void volk_gnsssdr_16ic_resampler_16ic_generic(lv_16sc_t* result, c
|
|||
#ifdef LV_HAVE_SSE2
|
||||
#include <emmintrin.h>
|
||||
|
||||
static inline void volk_gnsssdr_16ic_resampler_16ic_a_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, float rem_code_phase_chips, float code_phase_step_chips, int code_length_chips, unsigned int num_output_samples)//, int* scratch_buffer, float* scratch_buffer_float)
|
||||
static inline void volk_gnsssdr_16ic_resampler_fast_16ic_a_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, float rem_code_phase_chips, float code_phase_step_chips, int code_length_chips, unsigned int num_output_samples)//, int* scratch_buffer, float* scratch_buffer_float)
|
||||
{
|
||||
_MM_SET_ROUNDING_MODE (_MM_ROUND_NEAREST);//_MM_ROUND_NEAREST, _MM_ROUND_DOWN, _MM_ROUND_UP, _MM_ROUND_TOWARD_ZERO
|
||||
unsigned int number;
|
||||
|
|
@ -175,7 +175,7 @@ static inline void volk_gnsssdr_16ic_resampler_16ic_a_sse2(lv_16sc_t* result, co
|
|||
#ifdef LV_HAVE_SSE2
|
||||
#include <emmintrin.h>
|
||||
|
||||
static inline void volk_gnsssdr_16ic_resampler_16ic_u_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, float rem_code_phase_chips, float code_phase_step_chips, int code_length_chips, unsigned int num_output_samples)//, int* scratch_buffer, float* scratch_buffer_float)
|
||||
static inline void volk_gnsssdr_16ic_resampler_fast_16ic_u_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, float rem_code_phase_chips, float code_phase_step_chips, int code_length_chips, unsigned int num_output_samples)//, int* scratch_buffer, float* scratch_buffer_float)
|
||||
{
|
||||
_MM_SET_ROUNDING_MODE (_MM_ROUND_NEAREST);//_MM_ROUND_NEAREST, _MM_ROUND_DOWN, _MM_ROUND_UP, _MM_ROUND_TOWARD_ZERO
|
||||
unsigned int number;
|
||||
|
|
@ -257,7 +257,7 @@ static inline void volk_gnsssdr_16ic_resampler_16ic_u_sse2(lv_16sc_t* result, co
|
|||
#ifdef LV_HAVE_NEON
|
||||
#include <arm_neon.h>
|
||||
|
||||
static inline void volk_gnsssdr_16ic_resampler_16ic_neon(lv_16sc_t* result, const lv_16sc_t* local_code, float rem_code_phase_chips, float code_phase_step_chips, int code_length_chips, unsigned int num_output_samples)//, int* scratch_buffer, float* scratch_buffer_float)
|
||||
static inline void volk_gnsssdr_16ic_resampler_fast_16ic_neon(lv_16sc_t* result, const lv_16sc_t* local_code, float rem_code_phase_chips, float code_phase_step_chips, int code_length_chips, unsigned int num_output_samples)//, int* scratch_buffer, float* scratch_buffer_float)
|
||||
{
|
||||
unsigned int number;
|
||||
const unsigned int quarterPoints = num_output_samples / 4;
|
||||
|
|
@ -337,4 +337,4 @@ static inline void volk_gnsssdr_16ic_resampler_16ic_neon(lv_16sc_t* result, cons
|
|||
|
||||
#endif /* LV_HAVE_NEON */
|
||||
|
||||
#endif /*INCLUDED_volk_gnsssdr_16ic_resampler_16ic_H*/
|
||||
#endif /*INCLUDED_volk_gnsssdr_16ic_resampler_fast_16ic_H*/
|
||||
|
|
@ -1,5 +1,5 @@
|
|||
/*!
|
||||
* \file volk_gnsssdr_16ic_resamplerpuppet_16ic.h
|
||||
* \file volk_gnsssdr_16ic_resamplerfastpuppet_16ic.h
|
||||
* \brief VOLK_GNSSSDR puppet for the 16-bit complex vector resampler kernel.
|
||||
* \authors <ul>
|
||||
* <li> Carles Fernandez Prades 2016 cfernandez at cttc dot cat
|
||||
|
|
@ -32,56 +32,56 @@
|
|||
* -------------------------------------------------------------------------
|
||||
*/
|
||||
|
||||
#ifndef INCLUDED_volk_gnsssdr_16ic_resamplerpuppet_16ic_H
|
||||
#define INCLUDED_volk_gnsssdr_16ic_resamplerpuppet_16ic_H
|
||||
#ifndef INCLUDED_volk_gnsssdr_16ic_resamplerfastpuppet_16ic_H
|
||||
#define INCLUDED_volk_gnsssdr_16ic_resamplerfastpuppet_16ic_H
|
||||
|
||||
#include "volk_gnsssdr/volk_gnsssdr_16ic_resampler_16ic.h"
|
||||
#include "volk_gnsssdr/volk_gnsssdr_16ic_resampler_fast_16ic.h"
|
||||
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
static inline void volk_gnsssdr_16ic_resamplerpuppet_16ic_generic(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
static inline void volk_gnsssdr_16ic_resamplerfastpuppet_16ic_generic(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float rem_code_phase_chips = -0.123;
|
||||
float code_phase_step_chips = 0.1;
|
||||
int code_length_chips = 1023;
|
||||
volk_gnsssdr_16ic_resampler_16ic_generic(result, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_points);
|
||||
volk_gnsssdr_16ic_resampler_fast_16ic_generic(result, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_points);
|
||||
}
|
||||
|
||||
#endif /* LV_HAVE_GENERIC */
|
||||
|
||||
#ifdef LV_HAVE_SSE2
|
||||
static inline void volk_gnsssdr_16ic_resamplerpuppet_16ic_a_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
static inline void volk_gnsssdr_16ic_resamplerfastpuppet_16ic_a_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float rem_code_phase_chips = -0.123;
|
||||
float code_phase_step_chips = 0.1;
|
||||
int code_length_chips = 1023;
|
||||
volk_gnsssdr_16ic_resampler_16ic_a_sse2(result, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_points );
|
||||
volk_gnsssdr_16ic_resampler_fast_16ic_a_sse2(result, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_points );
|
||||
}
|
||||
|
||||
#endif /* LV_HAVE_SSE2 */
|
||||
|
||||
#ifdef LV_HAVE_SSE2
|
||||
|
||||
static inline void volk_gnsssdr_16ic_resamplerpuppet_16ic_u_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
static inline void volk_gnsssdr_16ic_resamplerfastpuppet_16ic_u_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float rem_code_phase_chips = -0.123;
|
||||
float code_phase_step_chips = 0.1;
|
||||
int code_length_chips = 1023;
|
||||
volk_gnsssdr_16ic_resampler_16ic_u_sse2(result, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_points );
|
||||
volk_gnsssdr_16ic_resampler_fast_16ic_u_sse2(result, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_points );
|
||||
}
|
||||
|
||||
#endif /* LV_HAVE_SSE2 */
|
||||
|
||||
#ifdef LV_HAVE_NEON
|
||||
|
||||
static inline void volk_gnsssdr_16ic_resamplerpuppet_16ic_neon(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
static inline void volk_gnsssdr_16ic_resamplerfastpuppet_16ic_neon(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float rem_code_phase_chips = -0.123;
|
||||
float code_phase_step_chips = 0.1;
|
||||
int code_length_chips = 1023;
|
||||
volk_gnsssdr_16ic_resampler_16ic_neon(result, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_points );
|
||||
volk_gnsssdr_16ic_resampler_fast_16ic_neon(result, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_points );
|
||||
}
|
||||
|
||||
#endif /* LV_HAVE_NEON */
|
||||
|
||||
#endif // INCLUDED_volk_gnsssdr_16ic_resamplerpuppet_16ic_H
|
||||
#endif // INCLUDED_volk_gnsssdr_16ic_resamplerfastpuppet_16ic_H
|
||||
|
|
@ -0,0 +1,153 @@
|
|||
/*!
|
||||
* \file volk_gnsssdr_16ic_resamplerfastxnpuppet_16ic.h
|
||||
* \brief VOLK_GNSSSDR puppet for the multiple 16-bit complex vector resampler kernel.
|
||||
* \authors <ul>
|
||||
* <li> Carles Fernandez Prades 2016 cfernandez at cttc dot cat
|
||||
* </ul>
|
||||
*
|
||||
* VOLK_GNSSSDR puppet for integrating the multiple resampler into the test system
|
||||
*
|
||||
* -------------------------------------------------------------------------
|
||||
*
|
||||
* Copyright (C) 2010-2015 (see AUTHORS file for a list of contributors)
|
||||
*
|
||||
* GNSS-SDR is a software defined Global Navigation
|
||||
* Satellite Systems receiver
|
||||
*
|
||||
* This file is part of GNSS-SDR.
|
||||
*
|
||||
* GNSS-SDR is free software: you can redistribute it and/or modify
|
||||
* it under the terms of the GNU General Public License as published by
|
||||
* the Free Software Foundation, either version 3 of the License, or
|
||||
* (at your option) any later version.
|
||||
*
|
||||
* GNSS-SDR is distributed in the hope that it will be useful,
|
||||
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
* GNU General Public License for more details.
|
||||
*
|
||||
* You should have received a copy of the GNU General Public License
|
||||
* along with GNSS-SDR. If not, see <http://www.gnu.org/licenses/>.
|
||||
*
|
||||
* -------------------------------------------------------------------------
|
||||
*/
|
||||
|
||||
#ifndef INCLUDED_volk_gnsssdr_16ic_resamplerfastxnpuppet_16ic_H
|
||||
#define INCLUDED_volk_gnsssdr_16ic_resamplerfastxnpuppet_16ic_H
|
||||
|
||||
#include "volk_gnsssdr/volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn.h"
|
||||
#include <volk_gnsssdr/volk_gnsssdr_malloc.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr.h>
|
||||
#include <string.h>
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
static inline void volk_gnsssdr_16ic_resamplerfastxnpuppet_16ic_generic(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = 0.1;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float* rem_code_phase_chips = (float*)volk_gnsssdr_malloc(sizeof(float) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
rem_code_phase_chips[n] = -0.234;
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn_generic(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
volk_gnsssdr_free(rem_code_phase_chips);
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif /* LV_HAVE_GENERIC */
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE2
|
||||
static inline void volk_gnsssdr_16ic_resamplerfastxnpuppet_16ic_a_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = 0.1;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float * rem_code_phase_chips = (float*)volk_gnsssdr_malloc(sizeof(float) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
rem_code_phase_chips[n] = -0.234;
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn_a_sse2(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy(result, result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
volk_gnsssdr_free(rem_code_phase_chips);
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE2
|
||||
static inline void volk_gnsssdr_16ic_resamplerfastxnpuppet_16ic_u_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = 0.1;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float * rem_code_phase_chips = (float*)volk_gnsssdr_malloc(sizeof(float) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
rem_code_phase_chips[n] = -0.234;
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn_u_sse2(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy(result, result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
volk_gnsssdr_free(rem_code_phase_chips);
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_NEON
|
||||
static inline void volk_gnsssdr_16ic_resamplerfastxnpuppet_16ic_neon(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = 0.1;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float * rem_code_phase_chips = (float*)volk_gnsssdr_malloc(sizeof(float) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
rem_code_phase_chips[n] = -0.234;
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn_neon(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy(result, result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
volk_gnsssdr_free(rem_code_phase_chips);
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#endif // INCLUDED_volk_gnsssdr_16ic_resamplerpuppet_16ic_H
|
||||
|
|
@ -1,282 +0,0 @@
|
|||
/*!
|
||||
* \file volk_gnsssdr_16ic_resamplerxnpuppet_16ic.h
|
||||
* \brief VOLK_GNSSSDR puppet for the multiple 16-bit complex vector resampler kernel.
|
||||
* \authors <ul>
|
||||
* <li> Carles Fernandez Prades 2016 cfernandez at cttc dot cat
|
||||
* </ul>
|
||||
*
|
||||
* VOLK_GNSSSDR puppet for integrating the multiple resampler into the test system
|
||||
*
|
||||
* -------------------------------------------------------------------------
|
||||
*
|
||||
* Copyright (C) 2010-2015 (see AUTHORS file for a list of contributors)
|
||||
*
|
||||
* GNSS-SDR is a software defined Global Navigation
|
||||
* Satellite Systems receiver
|
||||
*
|
||||
* This file is part of GNSS-SDR.
|
||||
*
|
||||
* GNSS-SDR is free software: you can redistribute it and/or modify
|
||||
* it under the terms of the GNU General Public License as published by
|
||||
* the Free Software Foundation, either version 3 of the License, or
|
||||
* (at your option) any later version.
|
||||
*
|
||||
* GNSS-SDR is distributed in the hope that it will be useful,
|
||||
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
* GNU General Public License for more details.
|
||||
*
|
||||
* You should have received a copy of the GNU General Public License
|
||||
* along with GNSS-SDR. If not, see <http://www.gnu.org/licenses/>.
|
||||
*
|
||||
* -------------------------------------------------------------------------
|
||||
*/
|
||||
|
||||
#ifndef INCLUDED_volk_gnsssdr_16ic_resamplerxnpuppet2_16ic_H
|
||||
#define INCLUDED_volk_gnsssdr_16ic_resamplerxnpuppet2_16ic_H
|
||||
|
||||
#include "volk_gnsssdr/volk_gnsssdr_16ic_xn_resampler2_16ic_xn.h"
|
||||
#include <volk_gnsssdr/volk_gnsssdr_malloc.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr.h>
|
||||
#include <string.h>
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
static inline void volk_gnsssdr_16ic_resamplerxnpuppet2_16ic_generic(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler2_16ic_xn_generic(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif /* LV_HAVE_GENERIC */
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE3
|
||||
static inline void volk_gnsssdr_16ic_resamplerxnpuppet2_16ic_a_sse3(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler2_16ic_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((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int 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_16ic_resamplerxnpuppet2_16ic_u_sse3(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler2_16ic_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((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
static inline void volk_gnsssdr_16ic_resamplerxnpuppet2_16ic_u_sse4_1(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler2_16ic_xn_u_sse4_1(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
static inline void volk_gnsssdr_16ic_resamplerxnpuppet2_16ic_a_sse4_1(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler2_16ic_xn_a_sse4_1(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int 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_16ic_resamplerxnpuppet2_16ic_u_avx(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler2_16ic_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((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int 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_16ic_resamplerxnpuppet2_16ic_a_avx(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler2_16ic_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((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_NEON
|
||||
static inline void volk_gnsssdr_16ic_resamplerxnpuppet2_16ic_neon(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler2_16ic_xn_neon(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
#endif // INCLUDED_volk_gnsssdr_16ic_resamplerpuppet_16ic_H
|
||||
|
|
@ -44,21 +44,23 @@
|
|||
#ifdef LV_HAVE_GENERIC
|
||||
static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_generic(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = 0.1;
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float* rem_code_phase_chips = (float*)volk_gnsssdr_malloc(sizeof(float) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
rem_code_phase_chips[n] = -0.234;
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
volk_gnsssdr_16ic_xn_resampler_16ic_xn_generic(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler_16ic_xn_generic(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
volk_gnsssdr_free(rem_code_phase_chips);
|
||||
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
|
|
@ -67,25 +69,57 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_generic(lv_16sc_t* r
|
|||
}
|
||||
|
||||
#endif /* LV_HAVE_GENERIC */
|
||||
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE2
|
||||
static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_a_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
#ifdef LV_HAVE_SSE3
|
||||
static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_a_sse3(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = 0.1;
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float * rem_code_phase_chips = (float*)volk_gnsssdr_malloc(sizeof(float) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
rem_code_phase_chips[n] = -0.234;
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
volk_gnsssdr_16ic_xn_resampler_16ic_xn_a_sse2(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy(result, result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
volk_gnsssdr_free(rem_code_phase_chips);
|
||||
volk_gnsssdr_16ic_xn_resampler_16ic_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((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int 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_16ic_resamplerxnpuppet_16ic_u_sse3(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler_16ic_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((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
|
|
@ -96,23 +130,116 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_a_sse2(lv_16sc_t* re
|
|||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE2
|
||||
static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_u_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_u_sse4_1(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = 0.1;
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float * rem_code_phase_chips = (float*)volk_gnsssdr_malloc(sizeof(float) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
rem_code_phase_chips[n] = -0.234;
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
volk_gnsssdr_16ic_xn_resampler_16ic_xn_u_sse2(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy(result, result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
volk_gnsssdr_free(rem_code_phase_chips);
|
||||
volk_gnsssdr_16ic_xn_resampler_16ic_xn_u_sse4_1(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_a_sse4_1(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler_16ic_xn_a_sse4_1(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int 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_16ic_resamplerxnpuppet_16ic_u_avx(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler_16ic_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((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int 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_16ic_resamplerxnpuppet_16ic_a_avx(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler_16ic_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((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
|
|
@ -126,20 +253,23 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_u_sse2(lv_16sc_t* re
|
|||
#ifdef LV_HAVE_NEON
|
||||
static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_neon(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = 0.1;
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
int num_out_vectors = 3;
|
||||
float * rem_code_phase_chips = (float*)volk_gnsssdr_malloc(sizeof(float) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
rem_code_phase_chips[n] = -0.234;
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
result_aux[n] = (lv_16sc_t*)volk_gnsssdr_malloc(sizeof(lv_16sc_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
volk_gnsssdr_16ic_xn_resampler_16ic_xn_neon(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy(result, result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
volk_gnsssdr_free(rem_code_phase_chips);
|
||||
volk_gnsssdr_16ic_xn_resampler_16ic_xn_neon(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((lv_16sc_t*)result, (lv_16sc_t*)result_aux[0], sizeof(lv_16sc_t) * num_points);
|
||||
|
||||
for(unsigned int n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
|
|
@ -149,5 +279,4 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_neon(lv_16sc_t* resu
|
|||
|
||||
#endif
|
||||
|
||||
|
||||
#endif // INCLUDED_volk_gnsssdr_16ic_resamplerpuppet_16ic_H
|
||||
|
|
|
|||
|
|
@ -1,591 +0,0 @@
|
|||
/*!
|
||||
* \file volk_gnsssdr_16ic_xn_resampler2_16ic_xn.h
|
||||
* \brief VOLK_GNSSSDR kernel: Resamples N 16 bits integer short complex vectors using zero hold resample algorithm.
|
||||
* \authors <ul>
|
||||
* <li> Javier Arribas, 2015. jarribas(at)cttc.es
|
||||
* </ul>
|
||||
*
|
||||
* VOLK_GNSSSDR kernel that resamples N 16 bits integer short complex vectors using zero hold resample algorithm.
|
||||
* It resamples a single GNSS local code signal replica into N vectors fractional-resampled and fractional-delayed
|
||||
* (i.e. it creates the Early, Prompt, and Late code replicas)
|
||||
*
|
||||
* -------------------------------------------------------------------------
|
||||
*
|
||||
* Copyright (C) 2010-2015 (see AUTHORS file for a list of contributors)
|
||||
*
|
||||
* GNSS-SDR is a software defined Global Navigation
|
||||
* Satellite Systems receiver
|
||||
*
|
||||
* This file is part of GNSS-SDR.
|
||||
*
|
||||
* GNSS-SDR is free software: you can redistribute it and/or modify
|
||||
* it under the terms of the GNU General Public License as published by
|
||||
* the Free Software Foundation, either version 3 of the License, or
|
||||
* (at your option) any later version.
|
||||
*
|
||||
* GNSS-SDR is distributed in the hope that it will be useful,
|
||||
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
* GNU General Public License for more details.
|
||||
*
|
||||
* You should have received a copy of the GNU General Public License
|
||||
* along with GNSS-SDR. If not, see <http://www.gnu.org/licenses/>.
|
||||
*
|
||||
* -------------------------------------------------------------------------
|
||||
*/
|
||||
|
||||
/*!
|
||||
* \page volk_gnsssdr_16ic_xn_resampler2_16ic_xn
|
||||
*
|
||||
* \b Overview
|
||||
*
|
||||
* Resamples a complex vector (16-bit integer each component), providing \p num_out_vectors outputs.
|
||||
*
|
||||
* <b>Dispatcher Prototype</b>
|
||||
* \code
|
||||
* void volk_gnsssdr_16ic_xn_resampler2_16ic_xn(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
* \endcode
|
||||
*
|
||||
* \b Inputs
|
||||
* \li local_code: One of the vectors to be multiplied.
|
||||
* \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.
|
||||
*
|
||||
* \b Outputs
|
||||
* \li result: Pointer to a vector of pointers where the results will be stored.
|
||||
*
|
||||
*/
|
||||
|
||||
#ifndef INCLUDED_volk_gnsssdr_16ic_xn_resampler2_16ic_xn_H
|
||||
#define INCLUDED_volk_gnsssdr_16ic_xn_resampler2_16ic_xn_H
|
||||
|
||||
#include <math.h>
|
||||
#include <stdlib.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_common.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
|
||||
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_generic(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
int local_code_chip_index;
|
||||
for (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
for (int 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[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 /*LV_HAVE_GENERIC*/
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
#include <smmintrin.h>
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_a_sse4_1(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
lv_16sc_t** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
|
||||
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, c, cTrunc, base;
|
||||
|
||||
for (int 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(unsigned int n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
aux = _mm_floor_ps(aux);
|
||||
|
||||
// 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(unsigned int 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(unsigned int 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_SSE4_1
|
||||
#include <smmintrin.h>
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_u_sse4_1(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
lv_16sc_t** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
|
||||
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, c, cTrunc, base;
|
||||
|
||||
for (int 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(unsigned int n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
aux = _mm_floor_ps(aux);
|
||||
|
||||
// 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(unsigned int 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(unsigned int 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_16ic_xn_resampler2_16ic_xn_a_sse3(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
lv_16sc_t** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
|
||||
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 (int 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(unsigned int 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(unsigned int 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(unsigned int 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_16ic_xn_resampler2_16ic_xn_u_sse3(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
lv_16sc_t** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
|
||||
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 (int 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(unsigned int 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(unsigned int 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(unsigned int 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_AVX
|
||||
#include <immintrin.h>
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_a_avx(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
lv_16sc_t** _result = result;
|
||||
const unsigned int avx_iters = num_points / 8;
|
||||
|
||||
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 (int 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(unsigned int n = 0; n < avx_iters; n++)
|
||||
{
|
||||
__builtin_prefetch(&_result[current_correlator_tap][8 * n + 7], 1, 0);
|
||||
__builtin_prefetch(&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(unsigned int 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 (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
for(unsigned int 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_16ic_xn_resampler2_16ic_xn_u_avx(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
lv_16sc_t** _result = result;
|
||||
const unsigned int avx_iters = num_points / 8;
|
||||
|
||||
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 (int 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(unsigned int n = 0; n < avx_iters; n++)
|
||||
{
|
||||
__builtin_prefetch(&_result[current_correlator_tap][8 * n + 7], 1, 0);
|
||||
__builtin_prefetch(&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(unsigned int 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 (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
for(unsigned int 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_NEON
|
||||
#include <arm_neon.h>
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_neon(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
lv_16sc_t** _result = result;
|
||||
const unsigned int neon_iters = num_points / 4;
|
||||
const int32x4_t ones = vdupq_n_s32(1);
|
||||
const float32x4_t fours = vdupq_n_f32(4.0f);
|
||||
const float32x4_t rem_code_phase_chips_reg = vdupq_n_f32(rem_code_phase_chips);
|
||||
const float32x4_t code_phase_step_chips_reg = vdupq_n_f32(code_phase_step_chips);
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) int32_t local_code_chip_index[4];
|
||||
int32_t local_code_chip_index_;
|
||||
|
||||
const int32x4_t zeros = vdupq_n_s32(0);
|
||||
const float32x4_t code_length_chips_reg_f = vdupq_n_f32((float)code_length_chips);
|
||||
const int32x4_t code_length_chips_reg_i = vdupq_n_s32((int32_t)code_length_chips);
|
||||
int32x4_t local_code_chip_index_reg, aux_i, negatives, i;
|
||||
float32x4_t aux, aux2, shifts_chips_reg, fi, c, j, cTrunc, base, indexn, reciprocal;
|
||||
__VOLK_ATTR_ALIGNED(16) const float vec[4] = { 0.0f, 1.0f, 2.0f, 3.0f };
|
||||
uint32x4_t igx;
|
||||
reciprocal = vrecpeq_f32(code_length_chips_reg_f);
|
||||
reciprocal = vmulq_f32(vrecpsq_f32(code_length_chips_reg_f, reciprocal), reciprocal);
|
||||
reciprocal = vmulq_f32(vrecpsq_f32(code_length_chips_reg_f, reciprocal), reciprocal); // this refinement is required!
|
||||
float32x4_t n0 = vld1q_f32((float*)vec);
|
||||
|
||||
for (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
shifts_chips_reg = vdupq_n_f32((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = vsubq_f32(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
indexn = n0;
|
||||
for(unsigned int n = 0; n < neon_iters; n++)
|
||||
{
|
||||
__builtin_prefetch(&_result[current_correlator_tap][4 * n + 3], 1, 0);
|
||||
__builtin_prefetch(&local_code_chip_index[4]);
|
||||
aux = vmulq_f32(code_phase_step_chips_reg, indexn);
|
||||
aux = vaddq_f32(aux, aux2);
|
||||
|
||||
//floor
|
||||
i = vcvtq_s32_f32(aux);
|
||||
fi = vcvtq_f32_s32(i);
|
||||
igx = vcgtq_f32(fi, aux);
|
||||
j = vcvtq_f32_s32(vandq_s32(vreinterpretq_s32_u32(igx), ones));
|
||||
aux = vsubq_f32(fi, j);
|
||||
|
||||
// fmod
|
||||
c = vmulq_f32(aux, reciprocal);
|
||||
i = vcvtq_s32_f32(c);
|
||||
cTrunc = vcvtq_f32_s32(i);
|
||||
base = vmulq_f32(cTrunc, code_length_chips_reg_f);
|
||||
aux = vsubq_f32(aux, base);
|
||||
local_code_chip_index_reg = vcvtq_s32_f32(aux);
|
||||
|
||||
negatives = vreinterpretq_s32_u32(vcltq_s32(local_code_chip_index_reg, zeros));
|
||||
aux_i = vandq_s32(code_length_chips_reg_i, negatives);
|
||||
local_code_chip_index_reg = vaddq_s32(local_code_chip_index_reg, aux_i);
|
||||
|
||||
vst1q_s32((int32_t*)local_code_chip_index, local_code_chip_index_reg);
|
||||
|
||||
for(unsigned int k = 0; k < 4; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = vaddq_f32(indexn, fours);
|
||||
}
|
||||
for(unsigned int n = neon_iters * 4; n < num_points; n++)
|
||||
{
|
||||
__builtin_prefetch(&_result[current_correlator_tap][n], 1, 0);
|
||||
// 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
|
||||
|
||||
|
||||
#endif /*INCLUDED_volk_gnsssdr_16ic_xn_resampler_16ic_xn_H*/
|
||||
|
||||
|
|
@ -5,8 +5,8 @@
|
|||
* <li> Javier Arribas, 2015. jarribas(at)cttc.es
|
||||
* </ul>
|
||||
*
|
||||
* VOLK_GNSSSDR kernel that esamples N 16 bits integer short complex vectors using zero hold resample algorithm.
|
||||
* It is optimized to resample a sigle GNSS local code signal replica into N vectors fractional-resampled and fractional-delayed
|
||||
* VOLK_GNSSSDR kernel that resamples N 16 bits integer short complex vectors using zero hold resample algorithm.
|
||||
* It resamples a single GNSS local code signal replica into N vectors fractional-resampled and fractional-delayed
|
||||
* (i.e. it creates the Early, Prompt, and Late code replicas)
|
||||
*
|
||||
* -------------------------------------------------------------------------
|
||||
|
|
@ -43,16 +43,17 @@
|
|||
*
|
||||
* <b>Dispatcher Prototype</b>
|
||||
* \code
|
||||
* void volk_gnsssdr_16ic_xn_resampler_16ic_xn(lv_16sc_t** result, const lv_16sc_t* local_code, float* rem_code_phase_chips, float code_phase_step_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_output_samples)
|
||||
* void volk_gnsssdr_16ic_xn_resampler_16ic_xn(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
* \endcode
|
||||
*
|
||||
* \b Inputs
|
||||
* \li local_code: One of the vectors to be multiplied.
|
||||
* \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_output_samples: The number of data values to be in the resampled vector.
|
||||
* \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.
|
||||
|
|
@ -63,26 +64,26 @@
|
|||
#define INCLUDED_volk_gnsssdr_16ic_xn_resampler_16ic_xn_H
|
||||
|
||||
#include <math.h>
|
||||
#include <stdlib.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_common.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
|
||||
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler_16ic_xn_generic(lv_16sc_t** result, const lv_16sc_t* local_code, float* rem_code_phase_chips, float code_phase_step_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_output_samples)
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler_16ic_xn_generic(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
int local_code_chip_index;
|
||||
//fesetround(FE_TONEAREST);
|
||||
for (int current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
for (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
for (unsigned int n = 0; n < num_output_samples; n++)
|
||||
for (int n = 0; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index = round(code_phase_step_chips * (float)(n) + rem_code_phase_chips[current_vector] - 0.5f);
|
||||
if (local_code_chip_index < 0.0) local_code_chip_index += code_length_chips;
|
||||
if (local_code_chip_index > (code_length_chips - 1)) local_code_chip_index -= code_length_chips;
|
||||
//std::cout<<"g["<<n<<"]="<<code_phase_step_chips*static_cast<float>(n) + rem_code_phase_chips-0.5f<<","<<local_code_chip_index<<" ";
|
||||
result[current_vector][n] = local_code[local_code_chip_index];
|
||||
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];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
|
@ -90,287 +91,501 @@ static inline void volk_gnsssdr_16ic_xn_resampler_16ic_xn_generic(lv_16sc_t** re
|
|||
#endif /*LV_HAVE_GENERIC*/
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE2
|
||||
#include <emmintrin.h>
|
||||
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler_16ic_xn_a_sse2(lv_16sc_t** result, const lv_16sc_t* local_code, float* rem_code_phase_chips ,float code_phase_step_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_output_samples)
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
#include <smmintrin.h>
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler_16ic_xn_a_sse4_1(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
_MM_SET_ROUNDING_MODE(_MM_ROUND_NEAREST);//_MM_ROUND_NEAREST, _MM_ROUND_DOWN, _MM_ROUND_UP, _MM_ROUND_TOWARD_ZERO
|
||||
unsigned int number;
|
||||
const unsigned int quarterPoints = num_output_samples / 4;
|
||||
|
||||
lv_16sc_t** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
|
||||
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];
|
||||
float tmp_rem_code_phase_chips;
|
||||
__m128 _rem_code_phase,_code_phase_step_chips;
|
||||
__m128i _code_length_chips,_code_length_chips_minus1;
|
||||
__m128 _code_phase_out,_code_phase_out_with_offset;
|
||||
int local_code_chip_index_;
|
||||
|
||||
_code_phase_step_chips = _mm_load1_ps(&code_phase_step_chips); //load float to all four float values in m128 register
|
||||
__VOLK_ATTR_ALIGNED(16) int four_times_code_length_chips_minus1[4];
|
||||
four_times_code_length_chips_minus1[0] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[1] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[2] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[3] = code_length_chips - 1;
|
||||
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, c, cTrunc, base;
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) int four_times_code_length_chips[4];
|
||||
four_times_code_length_chips[0] = code_length_chips;
|
||||
four_times_code_length_chips[1] = code_length_chips;
|
||||
four_times_code_length_chips[2] = code_length_chips;
|
||||
four_times_code_length_chips[3] = code_length_chips;
|
||||
|
||||
_code_length_chips = _mm_load_si128((__m128i*)&four_times_code_length_chips); //load float to all four float values in m128 register
|
||||
_code_length_chips_minus1 = _mm_load_si128((__m128i*)&four_times_code_length_chips_minus1); //load float to all four float values in m128 register
|
||||
|
||||
__m128i negative_indexes, overflow_indexes,_code_phase_out_int, _code_phase_out_int_neg,_code_phase_out_int_over;
|
||||
|
||||
__m128i zero = _mm_setzero_si128();
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) float init_idx_float[4] = { 0.0f, 1.0f, 2.0f, 3.0f };
|
||||
__m128 _4output_index = _mm_load_ps(init_idx_float);
|
||||
__VOLK_ATTR_ALIGNED(16) float init_4constant_float[4] = { 4.0f, 4.0f, 4.0f, 4.0f };
|
||||
__m128 _4constant_float = _mm_load_ps(init_4constant_float);
|
||||
|
||||
int current_vector = 0;
|
||||
int sample_idx = 0;
|
||||
for(number = 0; number < quarterPoints; number++)
|
||||
for (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
//common to all outputs
|
||||
_code_phase_out = _mm_mul_ps(_code_phase_step_chips, _4output_index); //compute the code phase point with the phase step
|
||||
|
||||
//output vector dependant (different code phase offset)
|
||||
for(current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
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(unsigned int n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
tmp_rem_code_phase_chips = rem_code_phase_chips[current_vector] - 0.5f; // adjust offset to perform correct rounding (chip transition at 0)
|
||||
_rem_code_phase = _mm_load1_ps(&tmp_rem_code_phase_chips); //load float to all four float values in m128 register
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
aux = _mm_floor_ps(aux);
|
||||
|
||||
_code_phase_out_with_offset = _mm_add_ps(_code_phase_out, _rem_code_phase); //add the phase offset
|
||||
_code_phase_out_int = _mm_cvtps_epi32(_code_phase_out_with_offset); //convert to integer
|
||||
// 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));
|
||||
|
||||
negative_indexes = _mm_cmplt_epi32(_code_phase_out_int, zero); //test for negative values
|
||||
_code_phase_out_int_neg = _mm_add_epi32(_code_phase_out_int, _code_length_chips); //the negative values branch
|
||||
_code_phase_out_int_neg = _mm_xor_si128(_code_phase_out_int, _mm_and_si128( negative_indexes, _mm_xor_si128( _code_phase_out_int_neg, _code_phase_out_int )));
|
||||
|
||||
overflow_indexes = _mm_cmpgt_epi32(_code_phase_out_int_neg, _code_length_chips_minus1); //test for overflow values
|
||||
_code_phase_out_int_over = _mm_sub_epi32(_code_phase_out_int_neg, _code_length_chips); //the negative values branch
|
||||
_code_phase_out_int_over = _mm_xor_si128(_code_phase_out_int_neg, _mm_and_si128( overflow_indexes, _mm_xor_si128( _code_phase_out_int_over, _code_phase_out_int_neg )));
|
||||
|
||||
_mm_store_si128((__m128i*)local_code_chip_index, _code_phase_out_int_over); // Store the results back
|
||||
|
||||
//todo: optimize the local code lookup table with intrinsics, if possible
|
||||
_result[current_vector][sample_idx] = local_code[local_code_chip_index[0]];
|
||||
_result[current_vector][sample_idx + 1] = local_code[local_code_chip_index[1]];
|
||||
_result[current_vector][sample_idx + 2] = local_code[local_code_chip_index[2]];
|
||||
_result[current_vector][sample_idx + 3] = local_code[local_code_chip_index[3]];
|
||||
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(unsigned int 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);
|
||||
}
|
||||
_4output_index = _mm_add_ps(_4output_index, _4constant_float);
|
||||
sample_idx += 4;
|
||||
}
|
||||
|
||||
for(number = quarterPoints * 4; number < num_output_samples; number++)
|
||||
{
|
||||
for(current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
for(unsigned int n = quarterPoints * 4; n < num_points; n++)
|
||||
{
|
||||
local_code_chip_index[0] = (int)(code_phase_step_chips * (float)(number) + rem_code_phase_chips[current_vector]);
|
||||
if (local_code_chip_index[0] < 0.0) local_code_chip_index[0] += code_length_chips - 1;
|
||||
if (local_code_chip_index[0] > (code_length_chips - 1)) local_code_chip_index[0] -= code_length_chips;
|
||||
_result[current_vector][number] = local_code[local_code_chip_index[0]];
|
||||
// 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 /* LV_HAVE_SSE2 */
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE2
|
||||
#include <emmintrin.h>
|
||||
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler_16ic_xn_u_sse2(lv_16sc_t** result, const lv_16sc_t* local_code, float* rem_code_phase_chips ,float code_phase_step_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_output_samples)
|
||||
{
|
||||
_MM_SET_ROUNDING_MODE(_MM_ROUND_NEAREST);//_MM_ROUND_NEAREST, _MM_ROUND_DOWN, _MM_ROUND_UP, _MM_ROUND_TOWARD_ZERO
|
||||
unsigned int number;
|
||||
const unsigned int quarterPoints = num_output_samples / 4;
|
||||
|
||||
lv_16sc_t** _result = result;
|
||||
__VOLK_ATTR_ALIGNED(16) int local_code_chip_index[4];
|
||||
float tmp_rem_code_phase_chips;
|
||||
__m128 _rem_code_phase,_code_phase_step_chips;
|
||||
__m128i _code_length_chips,_code_length_chips_minus1;
|
||||
__m128 _code_phase_out,_code_phase_out_with_offset;
|
||||
|
||||
_code_phase_step_chips = _mm_load1_ps(&code_phase_step_chips); //load float to all four float values in m128 register
|
||||
__VOLK_ATTR_ALIGNED(16) int four_times_code_length_chips_minus1[4];
|
||||
four_times_code_length_chips_minus1[0] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[1] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[2] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[3] = code_length_chips - 1;
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) int four_times_code_length_chips[4];
|
||||
four_times_code_length_chips[0] = code_length_chips;
|
||||
four_times_code_length_chips[1] = code_length_chips;
|
||||
four_times_code_length_chips[2] = code_length_chips;
|
||||
four_times_code_length_chips[3] = code_length_chips;
|
||||
|
||||
_code_length_chips = _mm_loadu_si128((__m128i*)&four_times_code_length_chips); //load float to all four float values in m128 register
|
||||
_code_length_chips_minus1 = _mm_loadu_si128((__m128i*)&four_times_code_length_chips_minus1); //load float to all four float values in m128 register
|
||||
|
||||
__m128i negative_indexes, overflow_indexes,_code_phase_out_int, _code_phase_out_int_neg,_code_phase_out_int_over;
|
||||
|
||||
__m128i zero = _mm_setzero_si128();
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) float init_idx_float[4] = { 0.0f, 1.0f, 2.0f, 3.0f };
|
||||
__m128 _4output_index = _mm_loadu_ps(init_idx_float);
|
||||
__VOLK_ATTR_ALIGNED(16) float init_4constant_float[4] = { 4.0f, 4.0f, 4.0f, 4.0f };
|
||||
__m128 _4constant_float = _mm_loadu_ps(init_4constant_float);
|
||||
|
||||
int current_vector = 0;
|
||||
int sample_idx = 0;
|
||||
for(number = 0; number < quarterPoints; number++)
|
||||
{
|
||||
//common to all outputs
|
||||
_code_phase_out = _mm_mul_ps(_code_phase_step_chips, _4output_index); //compute the code phase point with the phase step
|
||||
|
||||
//output vector dependant (different code phase offset)
|
||||
for(current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
{
|
||||
tmp_rem_code_phase_chips = rem_code_phase_chips[current_vector] - 0.5f; // adjust offset to perform correct rounding (chip transition at 0)
|
||||
_rem_code_phase = _mm_load1_ps(&tmp_rem_code_phase_chips); //load float to all four float values in m128 register
|
||||
|
||||
_code_phase_out_with_offset = _mm_add_ps(_code_phase_out, _rem_code_phase); //add the phase offset
|
||||
_code_phase_out_int = _mm_cvtps_epi32(_code_phase_out_with_offset); //convert to integer
|
||||
|
||||
negative_indexes = _mm_cmplt_epi32(_code_phase_out_int, zero); //test for negative values
|
||||
_code_phase_out_int_neg = _mm_add_epi32(_code_phase_out_int, _code_length_chips); //the negative values branch
|
||||
_code_phase_out_int_neg = _mm_xor_si128(_code_phase_out_int, _mm_and_si128( negative_indexes, _mm_xor_si128( _code_phase_out_int_neg, _code_phase_out_int )));
|
||||
|
||||
overflow_indexes = _mm_cmpgt_epi32(_code_phase_out_int_neg, _code_length_chips_minus1); //test for overflow values
|
||||
_code_phase_out_int_over = _mm_sub_epi32(_code_phase_out_int_neg, _code_length_chips); //the negative values branch
|
||||
_code_phase_out_int_over = _mm_xor_si128(_code_phase_out_int_neg, _mm_and_si128( overflow_indexes, _mm_xor_si128( _code_phase_out_int_over, _code_phase_out_int_neg )));
|
||||
|
||||
_mm_storeu_si128((__m128i*)local_code_chip_index, _code_phase_out_int_over); // Store the results back
|
||||
|
||||
//todo: optimize the local code lookup table with intrinsics, if possible
|
||||
_result[current_vector][sample_idx] = local_code[local_code_chip_index[0]];
|
||||
_result[current_vector][sample_idx + 1] = local_code[local_code_chip_index[1]];
|
||||
_result[current_vector][sample_idx + 2] = local_code[local_code_chip_index[2]];
|
||||
_result[current_vector][sample_idx + 3] = local_code[local_code_chip_index[3]];
|
||||
}
|
||||
_4output_index = _mm_add_ps(_4output_index, _4constant_float);
|
||||
sample_idx += 4;
|
||||
}
|
||||
|
||||
for(number = quarterPoints * 4; number < num_output_samples; number++)
|
||||
{
|
||||
for(current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
{
|
||||
local_code_chip_index[0] = (int)(code_phase_step_chips * (float)(number) + rem_code_phase_chips[current_vector]);
|
||||
if (local_code_chip_index[0] < 0.0) local_code_chip_index[0] += code_length_chips - 1;
|
||||
if (local_code_chip_index[0] > (code_length_chips - 1)) local_code_chip_index[0] -= code_length_chips;
|
||||
_result[current_vector][number] = local_code[local_code_chip_index[0]];
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
#endif /* LV_HAVE_SSE2 */
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
#include <smmintrin.h>
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler_16ic_xn_u_sse4_1(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
lv_16sc_t** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
|
||||
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, c, cTrunc, base;
|
||||
|
||||
for (int 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(unsigned int n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
aux = _mm_floor_ps(aux);
|
||||
|
||||
// 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(unsigned int 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(unsigned int 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_16ic_xn_resampler_16ic_xn_a_sse3(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
lv_16sc_t** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
|
||||
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 (int 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(unsigned int 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(unsigned int 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(unsigned int 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_16ic_xn_resampler_16ic_xn_u_sse3(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
lv_16sc_t** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
|
||||
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 (int 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(unsigned int 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(unsigned int 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(unsigned int 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_AVX
|
||||
#include <immintrin.h>
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler_16ic_xn_a_avx(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
lv_16sc_t** _result = result;
|
||||
const unsigned int avx_iters = num_points / 8;
|
||||
|
||||
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 (int 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(unsigned int n = 0; n < avx_iters; n++)
|
||||
{
|
||||
__builtin_prefetch(&_result[current_correlator_tap][8 * n + 7], 1, 0);
|
||||
__builtin_prefetch(&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(unsigned int 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 (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
for(unsigned int 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_16ic_xn_resampler_16ic_xn_u_avx(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
lv_16sc_t** _result = result;
|
||||
const unsigned int avx_iters = num_points / 8;
|
||||
|
||||
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 (int 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(unsigned int n = 0; n < avx_iters; n++)
|
||||
{
|
||||
__builtin_prefetch(&_result[current_correlator_tap][8 * n + 7], 1, 0);
|
||||
__builtin_prefetch(&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(unsigned int 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 (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
for(unsigned int 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_NEON
|
||||
#include <arm_neon.h>
|
||||
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler_16ic_xn_neon(lv_16sc_t** result, const lv_16sc_t* local_code, float* rem_code_phase_chips ,float code_phase_step_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_output_samples)
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler_16ic_xn_neon(lv_16sc_t** result, const lv_16sc_t* 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)
|
||||
{
|
||||
unsigned int number;
|
||||
const unsigned int quarterPoints = num_output_samples / 4;
|
||||
float32x4_t half = vdupq_n_f32(0.5f);
|
||||
|
||||
lv_16sc_t** _result = result;
|
||||
__VOLK_ATTR_ALIGNED(16) int local_code_chip_index[4];
|
||||
float tmp_rem_code_phase_chips;
|
||||
float32x4_t _rem_code_phase, _code_phase_step_chips;
|
||||
int32x4_t _code_length_chips, _code_length_chips_minus1;
|
||||
float32x4_t _code_phase_out, _code_phase_out_with_offset;
|
||||
float32x4_t sign, PlusHalf, Round;
|
||||
const unsigned int neon_iters = num_points / 4;
|
||||
const int32x4_t ones = vdupq_n_s32(1);
|
||||
const float32x4_t fours = vdupq_n_f32(4.0f);
|
||||
const float32x4_t rem_code_phase_chips_reg = vdupq_n_f32(rem_code_phase_chips);
|
||||
const float32x4_t code_phase_step_chips_reg = vdupq_n_f32(code_phase_step_chips);
|
||||
|
||||
_code_phase_step_chips = vld1q_dup_f32(&code_phase_step_chips); //load float to all four float values in float32x4_t register
|
||||
__VOLK_ATTR_ALIGNED(16) int four_times_code_length_chips_minus1[4];
|
||||
four_times_code_length_chips_minus1[0] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[1] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[2] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[3] = code_length_chips - 1;
|
||||
__VOLK_ATTR_ALIGNED(16) int32_t local_code_chip_index[4];
|
||||
int32_t local_code_chip_index_;
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) int four_times_code_length_chips[4];
|
||||
four_times_code_length_chips[0] = code_length_chips;
|
||||
four_times_code_length_chips[1] = code_length_chips;
|
||||
four_times_code_length_chips[2] = code_length_chips;
|
||||
four_times_code_length_chips[3] = code_length_chips;
|
||||
const int32x4_t zeros = vdupq_n_s32(0);
|
||||
const float32x4_t code_length_chips_reg_f = vdupq_n_f32((float)code_length_chips);
|
||||
const int32x4_t code_length_chips_reg_i = vdupq_n_s32((int32_t)code_length_chips);
|
||||
int32x4_t local_code_chip_index_reg, aux_i, negatives, i;
|
||||
float32x4_t aux, aux2, shifts_chips_reg, fi, c, j, cTrunc, base, indexn, reciprocal;
|
||||
__VOLK_ATTR_ALIGNED(16) const float vec[4] = { 0.0f, 1.0f, 2.0f, 3.0f };
|
||||
uint32x4_t igx;
|
||||
reciprocal = vrecpeq_f32(code_length_chips_reg_f);
|
||||
reciprocal = vmulq_f32(vrecpsq_f32(code_length_chips_reg_f, reciprocal), reciprocal);
|
||||
reciprocal = vmulq_f32(vrecpsq_f32(code_length_chips_reg_f, reciprocal), reciprocal); // this refinement is required!
|
||||
float32x4_t n0 = vld1q_f32((float*)vec);
|
||||
|
||||
_code_length_chips = vld1q_s32((int32_t*)&four_times_code_length_chips); //load float to all four float values in float32x4_t register
|
||||
_code_length_chips_minus1 = vld1q_s32((int32_t*)&four_times_code_length_chips_minus1); //load float to all four float values in float32x4_t register
|
||||
|
||||
int32x4_t _code_phase_out_int, _code_phase_out_int_neg, _code_phase_out_int_over;
|
||||
uint32x4_t negative_indexes, overflow_indexes;
|
||||
int32x4_t zero = vmovq_n_s32(0);
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) float init_idx_float[4] = { 0.0f, 1.0f, 2.0f, 3.0f };
|
||||
float32x4_t _4output_index = vld1q_f32(init_idx_float);
|
||||
__VOLK_ATTR_ALIGNED(16) float init_4constant_float[4] = { 4.0f, 4.0f, 4.0f, 4.0f };
|
||||
float32x4_t _4constant_float = vld1q_f32(init_4constant_float);
|
||||
|
||||
int current_vector = 0;
|
||||
int sample_idx = 0;
|
||||
for(number = 0; number < quarterPoints; number++)
|
||||
for (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
//common to all outputs
|
||||
_code_phase_out = vmulq_f32(_code_phase_step_chips, _4output_index); //compute the code phase point with the phase step
|
||||
|
||||
//output vector dependant (different code phase offset)
|
||||
for(current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
shifts_chips_reg = vdupq_n_f32((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = vsubq_f32(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
indexn = n0;
|
||||
for(unsigned int n = 0; n < neon_iters; n++)
|
||||
{
|
||||
tmp_rem_code_phase_chips = rem_code_phase_chips[current_vector] - 0.5f; // adjust offset to perform correct rounding (chip transition at 0)
|
||||
_rem_code_phase = vld1q_dup_f32(&tmp_rem_code_phase_chips); //load float to all four float values in float32x4_t register
|
||||
__builtin_prefetch(&_result[current_correlator_tap][4 * n + 3], 1, 0);
|
||||
__builtin_prefetch(&local_code_chip_index[4]);
|
||||
aux = vmulq_f32(code_phase_step_chips_reg, indexn);
|
||||
aux = vaddq_f32(aux, aux2);
|
||||
|
||||
_code_phase_out_with_offset = vaddq_f32(_code_phase_out, _rem_code_phase); //add the phase offset
|
||||
//_code_phase_out_int = _mm_cvtps_epi32(_code_phase_out_with_offset); //convert to integer
|
||||
sign = vcvtq_f32_u32((vshrq_n_u32(vreinterpretq_u32_f32(_code_phase_out_with_offset), 31)));
|
||||
PlusHalf = vaddq_f32(_code_phase_out_with_offset, half);
|
||||
Round = vsubq_f32(PlusHalf, sign);
|
||||
_code_phase_out_int = vcvtq_s32_f32(Round);
|
||||
//floor
|
||||
i = vcvtq_s32_f32(aux);
|
||||
fi = vcvtq_f32_s32(i);
|
||||
igx = vcgtq_f32(fi, aux);
|
||||
j = vcvtq_f32_s32(vandq_s32(vreinterpretq_s32_u32(igx), ones));
|
||||
aux = vsubq_f32(fi, j);
|
||||
|
||||
negative_indexes = vcltq_s32(_code_phase_out_int, zero); //test for negative values
|
||||
_code_phase_out_int_neg = vaddq_s32(_code_phase_out_int, _code_length_chips); //the negative values branch
|
||||
_code_phase_out_int_neg = veorq_s32(_code_phase_out_int, vandq_s32( (int32x4_t)negative_indexes, veorq_s32( _code_phase_out_int_neg, _code_phase_out_int )));
|
||||
// fmod
|
||||
c = vmulq_f32(aux, reciprocal);
|
||||
i = vcvtq_s32_f32(c);
|
||||
cTrunc = vcvtq_f32_s32(i);
|
||||
base = vmulq_f32(cTrunc, code_length_chips_reg_f);
|
||||
aux = vsubq_f32(aux, base);
|
||||
local_code_chip_index_reg = vcvtq_s32_f32(aux);
|
||||
|
||||
overflow_indexes = vcgtq_s32(_code_phase_out_int_neg, _code_length_chips_minus1); //test for overflow values
|
||||
_code_phase_out_int_over = vsubq_s32(_code_phase_out_int_neg, _code_length_chips); //the negative values branch
|
||||
_code_phase_out_int_over = veorq_s32(_code_phase_out_int_neg, vandq_s32( (int32x4_t)overflow_indexes, veorq_s32( _code_phase_out_int_over, _code_phase_out_int_neg )));
|
||||
negatives = vreinterpretq_s32_u32(vcltq_s32(local_code_chip_index_reg, zeros));
|
||||
aux_i = vandq_s32(code_length_chips_reg_i, negatives);
|
||||
local_code_chip_index_reg = vaddq_s32(local_code_chip_index_reg, aux_i);
|
||||
|
||||
vst1q_s32((int32_t*)local_code_chip_index, _code_phase_out_int_over); // Store the results back
|
||||
vst1q_s32((int32_t*)local_code_chip_index, local_code_chip_index_reg);
|
||||
|
||||
//todo: optimize the local code lookup table with intrinsics, if possible
|
||||
_result[current_vector][sample_idx] = local_code[local_code_chip_index[0]];
|
||||
_result[current_vector][sample_idx + 1] = local_code[local_code_chip_index[1]];
|
||||
_result[current_vector][sample_idx + 2] = local_code[local_code_chip_index[2]];
|
||||
_result[current_vector][sample_idx + 3] = local_code[local_code_chip_index[3]];
|
||||
for(unsigned int k = 0; k < 4; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = vaddq_f32(indexn, fours);
|
||||
}
|
||||
_4output_index = vaddq_f32(_4output_index, _4constant_float);
|
||||
sample_idx += 4;
|
||||
}
|
||||
|
||||
for(number = quarterPoints * 4; number < num_output_samples; number++)
|
||||
{
|
||||
for(current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
for(unsigned int n = neon_iters * 4; n < num_points; n++)
|
||||
{
|
||||
local_code_chip_index[0] = (int)(code_phase_step_chips * (float)(number) + rem_code_phase_chips[current_vector]);
|
||||
if (local_code_chip_index[0] < 0.0) local_code_chip_index[0] += code_length_chips - 1;
|
||||
if (local_code_chip_index[0] > (code_length_chips - 1)) local_code_chip_index[0] -= code_length_chips;
|
||||
_result[current_vector][number] = local_code[local_code_chip_index[0]];
|
||||
__builtin_prefetch(&_result[current_correlator_tap][n], 1, 0);
|
||||
// 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 /* LV_HAVE_NEON */
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#endif /*INCLUDED_volk_gnsssdr_16ic_xn_resampler_16ic_xn_H*/
|
||||
|
||||
|
|
|
|||
|
|
@ -0,0 +1,376 @@
|
|||
/*!
|
||||
* \file volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn.h
|
||||
* \brief VOLK_GNSSSDR kernel: Resamples N 16 bits integer short complex vectors using zero hold resample algorithm.
|
||||
* \authors <ul>
|
||||
* <li> Javier Arribas, 2015. jarribas(at)cttc.es
|
||||
* </ul>
|
||||
*
|
||||
* VOLK_GNSSSDR kernel that esamples N 16 bits integer short complex vectors using zero hold resample algorithm.
|
||||
* It is optimized to resample a sigle GNSS local code signal replica into N vectors fractional-resampled and fractional-delayed
|
||||
* (i.e. it creates the Early, Prompt, and Late code replicas)
|
||||
*
|
||||
* -------------------------------------------------------------------------
|
||||
*
|
||||
* Copyright (C) 2010-2015 (see AUTHORS file for a list of contributors)
|
||||
*
|
||||
* GNSS-SDR is a software defined Global Navigation
|
||||
* Satellite Systems receiver
|
||||
*
|
||||
* This file is part of GNSS-SDR.
|
||||
*
|
||||
* GNSS-SDR is free software: you can redistribute it and/or modify
|
||||
* it under the terms of the GNU General Public License as published by
|
||||
* the Free Software Foundation, either version 3 of the License, or
|
||||
* (at your option) any later version.
|
||||
*
|
||||
* GNSS-SDR is distributed in the hope that it will be useful,
|
||||
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
||||
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
||||
* GNU General Public License for more details.
|
||||
*
|
||||
* You should have received a copy of the GNU General Public License
|
||||
* along with GNSS-SDR. If not, see <http://www.gnu.org/licenses/>.
|
||||
*
|
||||
* -------------------------------------------------------------------------
|
||||
*/
|
||||
|
||||
/*!
|
||||
* \page volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn
|
||||
*
|
||||
* \b Overview
|
||||
*
|
||||
* Resamples a complex vector (16-bit integer each component), providing \p num_out_vectors outputs.
|
||||
*
|
||||
* <b>Dispatcher Prototype</b>
|
||||
* \code
|
||||
* void volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn(lv_16sc_t** result, const lv_16sc_t* local_code, float* rem_code_phase_chips, float code_phase_step_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_output_samples)
|
||||
* \endcode
|
||||
*
|
||||
* \b Inputs
|
||||
* \li local_code: One of the vectors to be multiplied.
|
||||
* \li rem_code_phase_chips: Remnant code phase [chips].
|
||||
* \li code_phase_step_chips: Phase increment per sample [chips/sample].
|
||||
* \li code_length_chips: Code length in chips.
|
||||
* \li num_out_vectors Number of output vectors.
|
||||
* \li num_output_samples: 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.
|
||||
*
|
||||
*/
|
||||
|
||||
#ifndef INCLUDED_volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn_H
|
||||
#define INCLUDED_volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn_H
|
||||
|
||||
#include <math.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_common.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
|
||||
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn_generic(lv_16sc_t** result, const lv_16sc_t* local_code, float* rem_code_phase_chips, float code_phase_step_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_output_samples)
|
||||
{
|
||||
int local_code_chip_index;
|
||||
//fesetround(FE_TONEAREST);
|
||||
for (int current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
{
|
||||
for (unsigned int n = 0; n < num_output_samples; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index = round(code_phase_step_chips * (float)(n) + rem_code_phase_chips[current_vector] - 0.5f);
|
||||
if (local_code_chip_index < 0.0) local_code_chip_index += code_length_chips;
|
||||
if (local_code_chip_index > (code_length_chips - 1)) local_code_chip_index -= code_length_chips;
|
||||
//std::cout<<"g["<<n<<"]="<<code_phase_step_chips*static_cast<float>(n) + rem_code_phase_chips-0.5f<<","<<local_code_chip_index<<" ";
|
||||
result[current_vector][n] = local_code[local_code_chip_index];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif /*LV_HAVE_GENERIC*/
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE2
|
||||
#include <emmintrin.h>
|
||||
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn_a_sse2(lv_16sc_t** result, const lv_16sc_t* local_code, float* rem_code_phase_chips ,float code_phase_step_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_output_samples)
|
||||
{
|
||||
_MM_SET_ROUNDING_MODE(_MM_ROUND_NEAREST);//_MM_ROUND_NEAREST, _MM_ROUND_DOWN, _MM_ROUND_UP, _MM_ROUND_TOWARD_ZERO
|
||||
unsigned int number;
|
||||
const unsigned int quarterPoints = num_output_samples / 4;
|
||||
|
||||
lv_16sc_t** _result = result;
|
||||
__VOLK_ATTR_ALIGNED(16) int local_code_chip_index[4];
|
||||
float tmp_rem_code_phase_chips;
|
||||
__m128 _rem_code_phase,_code_phase_step_chips;
|
||||
__m128i _code_length_chips,_code_length_chips_minus1;
|
||||
__m128 _code_phase_out,_code_phase_out_with_offset;
|
||||
|
||||
_code_phase_step_chips = _mm_load1_ps(&code_phase_step_chips); //load float to all four float values in m128 register
|
||||
__VOLK_ATTR_ALIGNED(16) int four_times_code_length_chips_minus1[4];
|
||||
four_times_code_length_chips_minus1[0] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[1] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[2] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[3] = code_length_chips - 1;
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) int four_times_code_length_chips[4];
|
||||
four_times_code_length_chips[0] = code_length_chips;
|
||||
four_times_code_length_chips[1] = code_length_chips;
|
||||
four_times_code_length_chips[2] = code_length_chips;
|
||||
four_times_code_length_chips[3] = code_length_chips;
|
||||
|
||||
_code_length_chips = _mm_load_si128((__m128i*)&four_times_code_length_chips); //load float to all four float values in m128 register
|
||||
_code_length_chips_minus1 = _mm_load_si128((__m128i*)&four_times_code_length_chips_minus1); //load float to all four float values in m128 register
|
||||
|
||||
__m128i negative_indexes, overflow_indexes,_code_phase_out_int, _code_phase_out_int_neg,_code_phase_out_int_over;
|
||||
|
||||
__m128i zero = _mm_setzero_si128();
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) float init_idx_float[4] = { 0.0f, 1.0f, 2.0f, 3.0f };
|
||||
__m128 _4output_index = _mm_load_ps(init_idx_float);
|
||||
__VOLK_ATTR_ALIGNED(16) float init_4constant_float[4] = { 4.0f, 4.0f, 4.0f, 4.0f };
|
||||
__m128 _4constant_float = _mm_load_ps(init_4constant_float);
|
||||
|
||||
int current_vector = 0;
|
||||
int sample_idx = 0;
|
||||
for(number = 0; number < quarterPoints; number++)
|
||||
{
|
||||
//common to all outputs
|
||||
_code_phase_out = _mm_mul_ps(_code_phase_step_chips, _4output_index); //compute the code phase point with the phase step
|
||||
|
||||
//output vector dependant (different code phase offset)
|
||||
for(current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
{
|
||||
tmp_rem_code_phase_chips = rem_code_phase_chips[current_vector] - 0.5f; // adjust offset to perform correct rounding (chip transition at 0)
|
||||
_rem_code_phase = _mm_load1_ps(&tmp_rem_code_phase_chips); //load float to all four float values in m128 register
|
||||
|
||||
_code_phase_out_with_offset = _mm_add_ps(_code_phase_out, _rem_code_phase); //add the phase offset
|
||||
_code_phase_out_int = _mm_cvtps_epi32(_code_phase_out_with_offset); //convert to integer
|
||||
|
||||
negative_indexes = _mm_cmplt_epi32(_code_phase_out_int, zero); //test for negative values
|
||||
_code_phase_out_int_neg = _mm_add_epi32(_code_phase_out_int, _code_length_chips); //the negative values branch
|
||||
_code_phase_out_int_neg = _mm_xor_si128(_code_phase_out_int, _mm_and_si128( negative_indexes, _mm_xor_si128( _code_phase_out_int_neg, _code_phase_out_int )));
|
||||
|
||||
overflow_indexes = _mm_cmpgt_epi32(_code_phase_out_int_neg, _code_length_chips_minus1); //test for overflow values
|
||||
_code_phase_out_int_over = _mm_sub_epi32(_code_phase_out_int_neg, _code_length_chips); //the negative values branch
|
||||
_code_phase_out_int_over = _mm_xor_si128(_code_phase_out_int_neg, _mm_and_si128( overflow_indexes, _mm_xor_si128( _code_phase_out_int_over, _code_phase_out_int_neg )));
|
||||
|
||||
_mm_store_si128((__m128i*)local_code_chip_index, _code_phase_out_int_over); // Store the results back
|
||||
|
||||
//todo: optimize the local code lookup table with intrinsics, if possible
|
||||
_result[current_vector][sample_idx] = local_code[local_code_chip_index[0]];
|
||||
_result[current_vector][sample_idx + 1] = local_code[local_code_chip_index[1]];
|
||||
_result[current_vector][sample_idx + 2] = local_code[local_code_chip_index[2]];
|
||||
_result[current_vector][sample_idx + 3] = local_code[local_code_chip_index[3]];
|
||||
}
|
||||
_4output_index = _mm_add_ps(_4output_index, _4constant_float);
|
||||
sample_idx += 4;
|
||||
}
|
||||
|
||||
for(number = quarterPoints * 4; number < num_output_samples; number++)
|
||||
{
|
||||
for(current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
{
|
||||
local_code_chip_index[0] = (int)(code_phase_step_chips * (float)(number) + rem_code_phase_chips[current_vector]);
|
||||
if (local_code_chip_index[0] < 0.0) local_code_chip_index[0] += code_length_chips - 1;
|
||||
if (local_code_chip_index[0] > (code_length_chips - 1)) local_code_chip_index[0] -= code_length_chips;
|
||||
_result[current_vector][number] = local_code[local_code_chip_index[0]];
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
}
|
||||
#endif /* LV_HAVE_SSE2 */
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE2
|
||||
#include <emmintrin.h>
|
||||
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn_u_sse2(lv_16sc_t** result, const lv_16sc_t* local_code, float* rem_code_phase_chips ,float code_phase_step_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_output_samples)
|
||||
{
|
||||
_MM_SET_ROUNDING_MODE(_MM_ROUND_NEAREST);//_MM_ROUND_NEAREST, _MM_ROUND_DOWN, _MM_ROUND_UP, _MM_ROUND_TOWARD_ZERO
|
||||
unsigned int number;
|
||||
const unsigned int quarterPoints = num_output_samples / 4;
|
||||
|
||||
lv_16sc_t** _result = result;
|
||||
__VOLK_ATTR_ALIGNED(16) int local_code_chip_index[4];
|
||||
float tmp_rem_code_phase_chips;
|
||||
__m128 _rem_code_phase,_code_phase_step_chips;
|
||||
__m128i _code_length_chips,_code_length_chips_minus1;
|
||||
__m128 _code_phase_out,_code_phase_out_with_offset;
|
||||
|
||||
_code_phase_step_chips = _mm_load1_ps(&code_phase_step_chips); //load float to all four float values in m128 register
|
||||
__VOLK_ATTR_ALIGNED(16) int four_times_code_length_chips_minus1[4];
|
||||
four_times_code_length_chips_minus1[0] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[1] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[2] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[3] = code_length_chips - 1;
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) int four_times_code_length_chips[4];
|
||||
four_times_code_length_chips[0] = code_length_chips;
|
||||
four_times_code_length_chips[1] = code_length_chips;
|
||||
four_times_code_length_chips[2] = code_length_chips;
|
||||
four_times_code_length_chips[3] = code_length_chips;
|
||||
|
||||
_code_length_chips = _mm_loadu_si128((__m128i*)&four_times_code_length_chips); //load float to all four float values in m128 register
|
||||
_code_length_chips_minus1 = _mm_loadu_si128((__m128i*)&four_times_code_length_chips_minus1); //load float to all four float values in m128 register
|
||||
|
||||
__m128i negative_indexes, overflow_indexes,_code_phase_out_int, _code_phase_out_int_neg,_code_phase_out_int_over;
|
||||
|
||||
__m128i zero = _mm_setzero_si128();
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) float init_idx_float[4] = { 0.0f, 1.0f, 2.0f, 3.0f };
|
||||
__m128 _4output_index = _mm_loadu_ps(init_idx_float);
|
||||
__VOLK_ATTR_ALIGNED(16) float init_4constant_float[4] = { 4.0f, 4.0f, 4.0f, 4.0f };
|
||||
__m128 _4constant_float = _mm_loadu_ps(init_4constant_float);
|
||||
|
||||
int current_vector = 0;
|
||||
int sample_idx = 0;
|
||||
for(number = 0; number < quarterPoints; number++)
|
||||
{
|
||||
//common to all outputs
|
||||
_code_phase_out = _mm_mul_ps(_code_phase_step_chips, _4output_index); //compute the code phase point with the phase step
|
||||
|
||||
//output vector dependant (different code phase offset)
|
||||
for(current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
{
|
||||
tmp_rem_code_phase_chips = rem_code_phase_chips[current_vector] - 0.5f; // adjust offset to perform correct rounding (chip transition at 0)
|
||||
_rem_code_phase = _mm_load1_ps(&tmp_rem_code_phase_chips); //load float to all four float values in m128 register
|
||||
|
||||
_code_phase_out_with_offset = _mm_add_ps(_code_phase_out, _rem_code_phase); //add the phase offset
|
||||
_code_phase_out_int = _mm_cvtps_epi32(_code_phase_out_with_offset); //convert to integer
|
||||
|
||||
negative_indexes = _mm_cmplt_epi32(_code_phase_out_int, zero); //test for negative values
|
||||
_code_phase_out_int_neg = _mm_add_epi32(_code_phase_out_int, _code_length_chips); //the negative values branch
|
||||
_code_phase_out_int_neg = _mm_xor_si128(_code_phase_out_int, _mm_and_si128( negative_indexes, _mm_xor_si128( _code_phase_out_int_neg, _code_phase_out_int )));
|
||||
|
||||
overflow_indexes = _mm_cmpgt_epi32(_code_phase_out_int_neg, _code_length_chips_minus1); //test for overflow values
|
||||
_code_phase_out_int_over = _mm_sub_epi32(_code_phase_out_int_neg, _code_length_chips); //the negative values branch
|
||||
_code_phase_out_int_over = _mm_xor_si128(_code_phase_out_int_neg, _mm_and_si128( overflow_indexes, _mm_xor_si128( _code_phase_out_int_over, _code_phase_out_int_neg )));
|
||||
|
||||
_mm_storeu_si128((__m128i*)local_code_chip_index, _code_phase_out_int_over); // Store the results back
|
||||
|
||||
//todo: optimize the local code lookup table with intrinsics, if possible
|
||||
_result[current_vector][sample_idx] = local_code[local_code_chip_index[0]];
|
||||
_result[current_vector][sample_idx + 1] = local_code[local_code_chip_index[1]];
|
||||
_result[current_vector][sample_idx + 2] = local_code[local_code_chip_index[2]];
|
||||
_result[current_vector][sample_idx + 3] = local_code[local_code_chip_index[3]];
|
||||
}
|
||||
_4output_index = _mm_add_ps(_4output_index, _4constant_float);
|
||||
sample_idx += 4;
|
||||
}
|
||||
|
||||
for(number = quarterPoints * 4; number < num_output_samples; number++)
|
||||
{
|
||||
for(current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
{
|
||||
local_code_chip_index[0] = (int)(code_phase_step_chips * (float)(number) + rem_code_phase_chips[current_vector]);
|
||||
if (local_code_chip_index[0] < 0.0) local_code_chip_index[0] += code_length_chips - 1;
|
||||
if (local_code_chip_index[0] > (code_length_chips - 1)) local_code_chip_index[0] -= code_length_chips;
|
||||
_result[current_vector][number] = local_code[local_code_chip_index[0]];
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
}
|
||||
|
||||
#endif /* LV_HAVE_SSE2 */
|
||||
|
||||
|
||||
#ifdef LV_HAVE_NEON
|
||||
#include <arm_neon.h>
|
||||
|
||||
static inline void volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn_neon(lv_16sc_t** result, const lv_16sc_t* local_code, float* rem_code_phase_chips ,float code_phase_step_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_output_samples)
|
||||
{
|
||||
unsigned int number;
|
||||
const unsigned int quarterPoints = num_output_samples / 4;
|
||||
float32x4_t half = vdupq_n_f32(0.5f);
|
||||
|
||||
lv_16sc_t** _result = result;
|
||||
__VOLK_ATTR_ALIGNED(16) int local_code_chip_index[4];
|
||||
float tmp_rem_code_phase_chips;
|
||||
float32x4_t _rem_code_phase, _code_phase_step_chips;
|
||||
int32x4_t _code_length_chips, _code_length_chips_minus1;
|
||||
float32x4_t _code_phase_out, _code_phase_out_with_offset;
|
||||
float32x4_t sign, PlusHalf, Round;
|
||||
|
||||
_code_phase_step_chips = vld1q_dup_f32(&code_phase_step_chips); //load float to all four float values in float32x4_t register
|
||||
__VOLK_ATTR_ALIGNED(16) int four_times_code_length_chips_minus1[4];
|
||||
four_times_code_length_chips_minus1[0] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[1] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[2] = code_length_chips - 1;
|
||||
four_times_code_length_chips_minus1[3] = code_length_chips - 1;
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) int four_times_code_length_chips[4];
|
||||
four_times_code_length_chips[0] = code_length_chips;
|
||||
four_times_code_length_chips[1] = code_length_chips;
|
||||
four_times_code_length_chips[2] = code_length_chips;
|
||||
four_times_code_length_chips[3] = code_length_chips;
|
||||
|
||||
_code_length_chips = vld1q_s32((int32_t*)&four_times_code_length_chips); //load float to all four float values in float32x4_t register
|
||||
_code_length_chips_minus1 = vld1q_s32((int32_t*)&four_times_code_length_chips_minus1); //load float to all four float values in float32x4_t register
|
||||
|
||||
int32x4_t _code_phase_out_int, _code_phase_out_int_neg, _code_phase_out_int_over;
|
||||
uint32x4_t negative_indexes, overflow_indexes;
|
||||
int32x4_t zero = vmovq_n_s32(0);
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) float init_idx_float[4] = { 0.0f, 1.0f, 2.0f, 3.0f };
|
||||
float32x4_t _4output_index = vld1q_f32(init_idx_float);
|
||||
__VOLK_ATTR_ALIGNED(16) float init_4constant_float[4] = { 4.0f, 4.0f, 4.0f, 4.0f };
|
||||
float32x4_t _4constant_float = vld1q_f32(init_4constant_float);
|
||||
|
||||
int current_vector = 0;
|
||||
int sample_idx = 0;
|
||||
for(number = 0; number < quarterPoints; number++)
|
||||
{
|
||||
//common to all outputs
|
||||
_code_phase_out = vmulq_f32(_code_phase_step_chips, _4output_index); //compute the code phase point with the phase step
|
||||
|
||||
//output vector dependant (different code phase offset)
|
||||
for(current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
{
|
||||
tmp_rem_code_phase_chips = rem_code_phase_chips[current_vector] - 0.5f; // adjust offset to perform correct rounding (chip transition at 0)
|
||||
_rem_code_phase = vld1q_dup_f32(&tmp_rem_code_phase_chips); //load float to all four float values in float32x4_t register
|
||||
|
||||
_code_phase_out_with_offset = vaddq_f32(_code_phase_out, _rem_code_phase); //add the phase offset
|
||||
//_code_phase_out_int = _mm_cvtps_epi32(_code_phase_out_with_offset); //convert to integer
|
||||
sign = vcvtq_f32_u32((vshrq_n_u32(vreinterpretq_u32_f32(_code_phase_out_with_offset), 31)));
|
||||
PlusHalf = vaddq_f32(_code_phase_out_with_offset, half);
|
||||
Round = vsubq_f32(PlusHalf, sign);
|
||||
_code_phase_out_int = vcvtq_s32_f32(Round);
|
||||
|
||||
negative_indexes = vcltq_s32(_code_phase_out_int, zero); //test for negative values
|
||||
_code_phase_out_int_neg = vaddq_s32(_code_phase_out_int, _code_length_chips); //the negative values branch
|
||||
_code_phase_out_int_neg = veorq_s32(_code_phase_out_int, vandq_s32( (int32x4_t)negative_indexes, veorq_s32( _code_phase_out_int_neg, _code_phase_out_int )));
|
||||
|
||||
overflow_indexes = vcgtq_s32(_code_phase_out_int_neg, _code_length_chips_minus1); //test for overflow values
|
||||
_code_phase_out_int_over = vsubq_s32(_code_phase_out_int_neg, _code_length_chips); //the negative values branch
|
||||
_code_phase_out_int_over = veorq_s32(_code_phase_out_int_neg, vandq_s32( (int32x4_t)overflow_indexes, veorq_s32( _code_phase_out_int_over, _code_phase_out_int_neg )));
|
||||
|
||||
vst1q_s32((int32_t*)local_code_chip_index, _code_phase_out_int_over); // Store the results back
|
||||
|
||||
//todo: optimize the local code lookup table with intrinsics, if possible
|
||||
_result[current_vector][sample_idx] = local_code[local_code_chip_index[0]];
|
||||
_result[current_vector][sample_idx + 1] = local_code[local_code_chip_index[1]];
|
||||
_result[current_vector][sample_idx + 2] = local_code[local_code_chip_index[2]];
|
||||
_result[current_vector][sample_idx + 3] = local_code[local_code_chip_index[3]];
|
||||
}
|
||||
_4output_index = vaddq_f32(_4output_index, _4constant_float);
|
||||
sample_idx += 4;
|
||||
}
|
||||
|
||||
for(number = quarterPoints * 4; number < num_output_samples; number++)
|
||||
{
|
||||
for(current_vector = 0; current_vector < num_out_vectors; current_vector++)
|
||||
{
|
||||
local_code_chip_index[0] = (int)(code_phase_step_chips * (float)(number) + rem_code_phase_chips[current_vector]);
|
||||
if (local_code_chip_index[0] < 0.0) local_code_chip_index[0] += code_length_chips - 1;
|
||||
if (local_code_chip_index[0] > (code_length_chips - 1)) local_code_chip_index[0] -= code_length_chips;
|
||||
_result[current_vector][number] = local_code[local_code_chip_index[0]];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif /* LV_HAVE_NEON */
|
||||
|
||||
#endif /*INCLUDED_volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn_H*/
|
||||
|
|
@ -85,9 +85,9 @@ std::vector<volk_gnsssdr_test_case_t> init_test_list(volk_gnsssdr_test_params_t
|
|||
(VOLK_INIT_TEST(volk_gnsssdr_16ic_convert_32fc, test_params_more_iters))
|
||||
(VOLK_INIT_PUPP(volk_gnsssdr_s32f_sincospuppet_32fc, volk_gnsssdr_s32f_sincos_32fc, test_params_inacc2))
|
||||
(VOLK_INIT_PUPP(volk_gnsssdr_16ic_rotatorpuppet_16ic, volk_gnsssdr_16ic_s32fc_x2_rotator_16ic, test_params_int1))
|
||||
(VOLK_INIT_PUPP(volk_gnsssdr_16ic_resamplerpuppet_16ic, volk_gnsssdr_16ic_resampler_16ic, test_params))
|
||||
(VOLK_INIT_PUPP(volk_gnsssdr_16ic_resamplerfastpuppet_16ic, volk_gnsssdr_16ic_resampler_fast_16ic, test_params))
|
||||
(VOLK_INIT_PUPP(volk_gnsssdr_16ic_resamplerfastxnpuppet_16ic, volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn, test_params))
|
||||
(VOLK_INIT_PUPP(volk_gnsssdr_16ic_resamplerxnpuppet_16ic, volk_gnsssdr_16ic_xn_resampler_16ic_xn, test_params))
|
||||
(VOLK_INIT_PUPP(volk_gnsssdr_16ic_resamplerxnpuppet2_16ic, volk_gnsssdr_16ic_xn_resampler2_16ic_xn, test_params))
|
||||
(VOLK_INIT_PUPP(volk_gnsssdr_32fc_resamplerxnpuppet_32fc, volk_gnsssdr_32fc_xn_resampler_32fc_xn, test_params))
|
||||
(VOLK_INIT_PUPP(volk_gnsssdr_16ic_x2_dotprodxnpuppet_16ic, volk_gnsssdr_16ic_x2_dot_prod_16ic_xn, test_params))
|
||||
(VOLK_INIT_PUPP(volk_gnsssdr_16ic_x2_rotator_dotprodxnpuppet_16ic, volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn, test_params_int16))
|
||||
|
|
|
|||
|
|
@ -85,7 +85,7 @@ void cpu_multicorrelator_16sc::update_local_code(int correlator_length_samples,
|
|||
d_tmp_code_phases_chips[n] = d_shifts_chips[n] - rem_code_phase_chips;
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_xn_resampler_16ic_xn(d_local_codes_resampled,
|
||||
volk_gnsssdr_16ic_xn_resampler_fast_16ic_xn(d_local_codes_resampled,
|
||||
d_local_code_in,
|
||||
d_tmp_code_phases_chips,
|
||||
code_phase_step_chips,
|
||||
|
|
|
|||
Loading…
Reference in New Issue
Block a user