mirror of
https://github.com/gnss-sdr/gnss-sdr
synced 2024-06-24 22:13:15 +00:00
Updated volk_gnsssdr_module for real codes
Added 16i and 32f resamplers and 32fc_32f and 16ic_16i rotator dot product to enable use of real (rather than complex) local code replicas
This commit is contained in:
Cillian O'Driscoll
parent
3f557eeb41
commit
676c1506da
|
|
@ -61,6 +61,14 @@ _mm256_magnitudesquared_ps(__m256 cplxValue1, __m256 cplxValue2){
|
|||
return _mm256_hadd_ps(complex1, complex2); // Add the I2 and Q2 values
|
||||
}
|
||||
|
||||
static inline __m256 _mm256_complexnormalise_ps( __m256 z ){
|
||||
__m256 tmp1 = _mm256_mul_ps(z, z);
|
||||
__m256 tmp2 = _mm256_hadd_ps(tmp1, tmp1);
|
||||
tmp1 = _mm256_shuffle_ps(tmp2, tmp2, 0xD8);
|
||||
tmp2 = _mm256_sqrt_ps(tmp1);
|
||||
return _mm256_div_ps(z, tmp2);
|
||||
}
|
||||
|
||||
static inline __m256
|
||||
_mm256_magnitude_ps(__m256 cplxValue1, __m256 cplxValue2){
|
||||
return _mm256_sqrt_ps(_mm256_magnitudesquared_ps(cplxValue1, cplxValue2));
|
||||
|
|
|
|||
|
|
@ -0,0 +1,282 @@
|
|||
/*!
|
||||
* \file volk_gnsssdr_16i_resamplerxnpuppet_16i.h
|
||||
* \brief VOLK_GNSSSDR puppet for the multiple 16-bit vector resampler kernel.
|
||||
* \authors <ul>
|
||||
* <li> Cillian O'Driscoll 2017 cillian.odriscoll at gmail dot com
|
||||
* </ul>
|
||||
*
|
||||
* VOLK_GNSSSDR puppet for integrating the multiple resampler into the test system
|
||||
*
|
||||
* -------------------------------------------------------------------------
|
||||
*
|
||||
* Copyright (C) 2010-2017 (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_16i_resamplerxnpuppet_16i_H
|
||||
#define INCLUDED_volk_gnsssdr_16i_resamplerxnpuppet_16i_H
|
||||
|
||||
#include "volk_gnsssdr/volk_gnsssdr_16i_xn_resampler_16i_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_16i_resamplerxnpuppet_16i_generic(int16_t* result, const int16_t* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
unsigned int n;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
int16_t** result_aux = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16i_xn_resampler_16i_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((int16_t*)result, (int16_t*)result_aux[0], sizeof(int16_t) * num_points);
|
||||
|
||||
for(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_16i_resamplerxnpuppet_16i_a_sse3(int16_t* result, const int16_t* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
int16_t** result_aux = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16i_xn_resampler_16i_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((int16_t*)result, (int16_t*)result_aux[0], sizeof(int16_t) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
#ifdef LV_HAVE_SSE3
|
||||
static inline void volk_gnsssdr_16i_resamplerxnpuppet_16i_u_sse3(int16_t* result, const int16_t* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
int16_t** result_aux = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16i_xn_resampler_16i_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((int16_t*)result, (int16_t*)result_aux[0], sizeof(int16_t) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
static inline void volk_gnsssdr_16i_resamplerxnpuppet_16i_u_sse4_1(int16_t* result, const int16_t* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
int16_t** result_aux = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16i_xn_resampler_16i_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((int16_t*)result, (int16_t*)result_aux[0], sizeof(int16_t) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
static inline void volk_gnsssdr_16i_resamplerxnpuppet_16i_a_sse4_1(int16_t* result, const int16_t* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
int16_t** result_aux = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16i_xn_resampler_16i_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((int16_t*)result, (int16_t*)result_aux[0], sizeof(int16_t) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_AVX
|
||||
static inline void volk_gnsssdr_16i_resamplerxnpuppet_16i_u_avx(int16_t* result, const int16_t* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
int16_t** result_aux = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16i_xn_resampler_16i_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((int16_t*)result, (int16_t*)result_aux[0], sizeof(int16_t) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_AVX
|
||||
static inline void volk_gnsssdr_16i_resamplerxnpuppet_16i_a_avx(int16_t* result, const int16_t* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
int16_t** result_aux = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16i_xn_resampler_16i_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((int16_t*)result, (int16_t*)result_aux[0], sizeof(int16_t) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_NEON
|
||||
static inline void volk_gnsssdr_16i_resamplerxnpuppet_16i_neon(int16_t* result, const int16_t* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
int16_t** result_aux = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_16i_xn_resampler_16i_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((int16_t*)result, (int16_t*)result_aux[0], sizeof(int16_t) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
#endif // INCLUDED_volk_gnsssdr_16i_resamplerpuppet_16i_H
|
||||
|
||||
|
|
@ -0,0 +1,608 @@
|
|||
/*!
|
||||
* \file volk_gnsssdr_16i_xn_resampler_16i_xn.h
|
||||
* \brief VOLK_GNSSSDR kernel: Resamples N 16 bits integer short vectors using zero hold resample algorithm.
|
||||
* \authors <ul>
|
||||
* <li> Cillian O'Driscoll, 2017. cillian.odriscoll(at)gmail.com
|
||||
* </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) 2010-2017 (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_16i_xn_resampler_16i_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_16i_xn_resampler_16i_xn(int16_t** result, const int16_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: Vector to be resampled.
|
||||
* \li rem_code_phase_chips: Remnant code phase [chips].
|
||||
* \li code_phase_step_chips: Phase increment per sample [chips/sample].
|
||||
* \li shifts_chips: Vector of floats that defines the spacing (in chips) between the replicas of \p local_code
|
||||
* \li code_length_chips: Code length in chips.
|
||||
* \li num_out_vectors: Number of output vectors.
|
||||
* \li num_points: The number of data values to be in the resampled vector.
|
||||
*
|
||||
* \b Outputs
|
||||
* \li result: Pointer to a vector of pointers where the results will be stored.
|
||||
*
|
||||
*/
|
||||
|
||||
#ifndef INCLUDED_volk_gnsssdr_16i_xn_resampler_16i_xn_H
|
||||
#define INCLUDED_volk_gnsssdr_16i_xn_resampler_16i_xn_H
|
||||
|
||||
#include <assert.h>
|
||||
#include <math.h>
|
||||
#include <stdlib.h>
|
||||
#include <stdio.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_common.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
|
||||
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
|
||||
static inline void volk_gnsssdr_16i_xn_resampler_16i_xn_generic(int16_t** result, const int16_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;
|
||||
int current_correlator_tap;
|
||||
int n;
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
for (n = 0; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index = (int)floor(code_phase_step_chips * (float)n + shifts_chips[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_16i_xn_resampler_16i_xn_a_sse4_1(int16_t** result, const int16_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)
|
||||
{
|
||||
int16_t** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
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 (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
shifts_chips_reg = _mm_set_ps1((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
__m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
|
||||
for(n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
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(k = 0; k < 4; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm_add_ps(indexn, fours);
|
||||
}
|
||||
for(n = quarterPoints * 4; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
|
||||
//Take into account that in multitap correlators, the shifts can be negative!
|
||||
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1);
|
||||
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
#include <smmintrin.h>
|
||||
static inline void volk_gnsssdr_16i_xn_resampler_16i_xn_u_sse4_1(int16_t** result, const int16_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)
|
||||
{
|
||||
int16_t** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
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 (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
shifts_chips_reg = _mm_set_ps1((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
__m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
|
||||
for(n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
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(k = 0; k < 4; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm_add_ps(indexn, fours);
|
||||
}
|
||||
for(n = quarterPoints * 4; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
|
||||
//Take into account that in multitap correlators, the shifts can be negative!
|
||||
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1);
|
||||
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE3
|
||||
#include <pmmintrin.h>
|
||||
static inline void volk_gnsssdr_16i_xn_resampler_16i_xn_a_sse3(int16_t** result, const int16_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)
|
||||
{
|
||||
int16_t** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
const __m128 ones = _mm_set1_ps(1.0f);
|
||||
const __m128 fours = _mm_set1_ps(4.0f);
|
||||
const __m128 rem_code_phase_chips_reg = _mm_set_ps1(rem_code_phase_chips);
|
||||
const __m128 code_phase_step_chips_reg = _mm_set_ps1(code_phase_step_chips);
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) int local_code_chip_index[4];
|
||||
int local_code_chip_index_;
|
||||
|
||||
const __m128i zeros = _mm_setzero_si128();
|
||||
const __m128 code_length_chips_reg_f = _mm_set_ps1((float)code_length_chips);
|
||||
const __m128i code_length_chips_reg_i = _mm_set1_epi32((int)code_length_chips);
|
||||
__m128i local_code_chip_index_reg, aux_i, negatives, i;
|
||||
__m128 aux, aux2, shifts_chips_reg, fi, igx, j, c, cTrunc, base;
|
||||
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
shifts_chips_reg = _mm_set_ps1((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
__m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
|
||||
for(n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
i = _mm_cvttps_epi32(aux);
|
||||
fi = _mm_cvtepi32_ps(i);
|
||||
igx = _mm_cmpgt_ps(fi, aux);
|
||||
j = _mm_and_ps(igx, ones);
|
||||
aux = _mm_sub_ps(fi, j);
|
||||
// fmod
|
||||
c = _mm_div_ps(aux, code_length_chips_reg_f);
|
||||
i = _mm_cvttps_epi32(c);
|
||||
cTrunc = _mm_cvtepi32_ps(i);
|
||||
base = _mm_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||
local_code_chip_index_reg = _mm_cvtps_epi32(_mm_sub_ps(aux, base));
|
||||
|
||||
negatives = _mm_cmplt_epi32(local_code_chip_index_reg, zeros);
|
||||
aux_i = _mm_and_si128(code_length_chips_reg_i, negatives);
|
||||
local_code_chip_index_reg = _mm_add_epi32(local_code_chip_index_reg, aux_i);
|
||||
_mm_store_si128((__m128i*)local_code_chip_index, local_code_chip_index_reg);
|
||||
for(k = 0; k < 4; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm_add_ps(indexn, fours);
|
||||
}
|
||||
for(n = quarterPoints * 4; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
|
||||
//Take into account that in multitap correlators, the shifts can be negative!
|
||||
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1);
|
||||
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE3
|
||||
#include <pmmintrin.h>
|
||||
static inline void volk_gnsssdr_16i_xn_resampler_16i_xn_u_sse3(int16_t** result, const int16_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)
|
||||
{
|
||||
int16_t** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
const __m128 ones = _mm_set1_ps(1.0f);
|
||||
const __m128 fours = _mm_set1_ps(4.0f);
|
||||
const __m128 rem_code_phase_chips_reg = _mm_set_ps1(rem_code_phase_chips);
|
||||
const __m128 code_phase_step_chips_reg = _mm_set_ps1(code_phase_step_chips);
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) int local_code_chip_index[4];
|
||||
int local_code_chip_index_;
|
||||
|
||||
const __m128i zeros = _mm_setzero_si128();
|
||||
const __m128 code_length_chips_reg_f = _mm_set_ps1((float)code_length_chips);
|
||||
const __m128i code_length_chips_reg_i = _mm_set1_epi32((int)code_length_chips);
|
||||
__m128i local_code_chip_index_reg, aux_i, negatives, i;
|
||||
__m128 aux, aux2, shifts_chips_reg, fi, igx, j, c, cTrunc, base;
|
||||
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
shifts_chips_reg = _mm_set_ps1((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
__m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
|
||||
for(n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
i = _mm_cvttps_epi32(aux);
|
||||
fi = _mm_cvtepi32_ps(i);
|
||||
igx = _mm_cmpgt_ps(fi, aux);
|
||||
j = _mm_and_ps(igx, ones);
|
||||
aux = _mm_sub_ps(fi, j);
|
||||
// fmod
|
||||
c = _mm_div_ps(aux, code_length_chips_reg_f);
|
||||
i = _mm_cvttps_epi32(c);
|
||||
cTrunc = _mm_cvtepi32_ps(i);
|
||||
base = _mm_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||
local_code_chip_index_reg = _mm_cvtps_epi32(_mm_sub_ps(aux, base));
|
||||
|
||||
negatives = _mm_cmplt_epi32(local_code_chip_index_reg, zeros);
|
||||
aux_i = _mm_and_si128(code_length_chips_reg_i, negatives);
|
||||
local_code_chip_index_reg = _mm_add_epi32(local_code_chip_index_reg, aux_i);
|
||||
_mm_store_si128((__m128i*)local_code_chip_index, local_code_chip_index_reg);
|
||||
for(k = 0; k < 4; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm_add_ps(indexn, fours);
|
||||
}
|
||||
for(n = quarterPoints * 4; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
|
||||
//Take into account that in multitap correlators, the shifts can be negative!
|
||||
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1);
|
||||
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_AVX
|
||||
#include <immintrin.h>
|
||||
static inline void volk_gnsssdr_16i_xn_resampler_16i_xn_a_avx(int16_t** result, const int16_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)
|
||||
{
|
||||
int16_t** _result = result;
|
||||
const unsigned int avx_iters = num_points / 8;
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
const __m256 eights = _mm256_set1_ps(8.0f);
|
||||
const __m256 rem_code_phase_chips_reg = _mm256_set1_ps(rem_code_phase_chips);
|
||||
const __m256 code_phase_step_chips_reg = _mm256_set1_ps(code_phase_step_chips);
|
||||
|
||||
__VOLK_ATTR_ALIGNED(32) int local_code_chip_index[8];
|
||||
int local_code_chip_index_;
|
||||
|
||||
const __m256 zeros = _mm256_setzero_ps();
|
||||
const __m256 code_length_chips_reg_f = _mm256_set1_ps((float)code_length_chips);
|
||||
const __m256 n0 = _mm256_set_ps(7.0f, 6.0f, 5.0f, 4.0f, 3.0f, 2.0f, 1.0f, 0.0f);
|
||||
|
||||
__m256i local_code_chip_index_reg, i;
|
||||
__m256 aux, aux2, aux3, shifts_chips_reg, c, cTrunc, base, negatives, indexn;
|
||||
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
shifts_chips_reg = _mm256_set1_ps((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = _mm256_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
indexn = n0;
|
||||
for(n = 0; n < avx_iters; n++)
|
||||
{
|
||||
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&_result[current_correlator_tap][8 * n + 7], 1, 0);
|
||||
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&local_code_chip_index[8], 1, 3);
|
||||
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm256_add_ps(aux, aux2);
|
||||
// floor
|
||||
aux = _mm256_floor_ps(aux);
|
||||
|
||||
// fmod
|
||||
c = _mm256_div_ps(aux, code_length_chips_reg_f);
|
||||
i = _mm256_cvttps_epi32(c);
|
||||
cTrunc = _mm256_cvtepi32_ps(i);
|
||||
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||
local_code_chip_index_reg = _mm256_cvttps_epi32(_mm256_sub_ps(aux, base));
|
||||
|
||||
// no negatives
|
||||
c = _mm256_cvtepi32_ps(local_code_chip_index_reg);
|
||||
negatives = _mm256_cmp_ps(c, zeros, 0x01 );
|
||||
aux3 = _mm256_and_ps(code_length_chips_reg_f, negatives);
|
||||
aux = _mm256_add_ps(c, aux3);
|
||||
local_code_chip_index_reg = _mm256_cvttps_epi32(aux);
|
||||
|
||||
_mm256_store_si256((__m256i*)local_code_chip_index, local_code_chip_index_reg);
|
||||
for(k = 0; k < 8; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 8 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm256_add_ps(indexn, eights);
|
||||
}
|
||||
}
|
||||
_mm256_zeroupper();
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
for(n = avx_iters * 8; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
|
||||
//Take into account that in multitap correlators, the shifts can be negative!
|
||||
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1);
|
||||
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_AVX
|
||||
#include <immintrin.h>
|
||||
static inline void volk_gnsssdr_16i_xn_resampler_16i_xn_u_avx(int16_t** result, const int16_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)
|
||||
{
|
||||
int16_t** _result = result;
|
||||
const unsigned int avx_iters = num_points / 8;
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
const __m256 eights = _mm256_set1_ps(8.0f);
|
||||
const __m256 rem_code_phase_chips_reg = _mm256_set1_ps(rem_code_phase_chips);
|
||||
const __m256 code_phase_step_chips_reg = _mm256_set1_ps(code_phase_step_chips);
|
||||
|
||||
__VOLK_ATTR_ALIGNED(32) int local_code_chip_index[8];
|
||||
int local_code_chip_index_;
|
||||
|
||||
const __m256 zeros = _mm256_setzero_ps();
|
||||
const __m256 code_length_chips_reg_f = _mm256_set1_ps((float)code_length_chips);
|
||||
const __m256 n0 = _mm256_set_ps(7.0f, 6.0f, 5.0f, 4.0f, 3.0f, 2.0f, 1.0f, 0.0f);
|
||||
|
||||
__m256i local_code_chip_index_reg, i;
|
||||
__m256 aux, aux2, aux3, shifts_chips_reg, c, cTrunc, base, negatives, indexn;
|
||||
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
shifts_chips_reg = _mm256_set1_ps((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = _mm256_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
indexn = n0;
|
||||
for(n = 0; n < avx_iters; n++)
|
||||
{
|
||||
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&_result[current_correlator_tap][8 * n + 7], 1, 0);
|
||||
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&local_code_chip_index[8], 1, 3);
|
||||
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm256_add_ps(aux, aux2);
|
||||
// floor
|
||||
aux = _mm256_floor_ps(aux);
|
||||
|
||||
// fmod
|
||||
c = _mm256_div_ps(aux, code_length_chips_reg_f);
|
||||
i = _mm256_cvttps_epi32(c);
|
||||
cTrunc = _mm256_cvtepi32_ps(i);
|
||||
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||
local_code_chip_index_reg = _mm256_cvttps_epi32(_mm256_sub_ps(aux, base));
|
||||
|
||||
// no negatives
|
||||
c = _mm256_cvtepi32_ps(local_code_chip_index_reg);
|
||||
negatives = _mm256_cmp_ps(c, zeros, 0x01 );
|
||||
aux3 = _mm256_and_ps(code_length_chips_reg_f, negatives);
|
||||
aux = _mm256_add_ps(c, aux3);
|
||||
local_code_chip_index_reg = _mm256_cvttps_epi32(aux);
|
||||
|
||||
_mm256_store_si256((__m256i*)local_code_chip_index, local_code_chip_index_reg);
|
||||
for(k = 0; k < 8; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 8 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm256_add_ps(indexn, eights);
|
||||
}
|
||||
}
|
||||
_mm256_zeroupper();
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
for(n = avx_iters * 8; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
|
||||
//Take into account that in multitap correlators, the shifts can be negative!
|
||||
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1);
|
||||
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_NEON
|
||||
#include <arm_neon.h>
|
||||
static inline void volk_gnsssdr_16i_xn_resampler_16i_xn_neon(int16_t** result, const int16_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)
|
||||
{
|
||||
int16_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);
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
for (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(n = 0; n < neon_iters; n++)
|
||||
{
|
||||
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&_result[current_correlator_tap][4 * n + 3], 1, 0);
|
||||
__VOLK_GNSSSDR_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(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(n = neon_iters * 4; n < num_points; n++)
|
||||
{
|
||||
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&_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_16i_xn_resampler_16i_xn_H*/
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
|
|
@ -0,0 +1,384 @@
|
|||
/*!
|
||||
* \file volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic.h
|
||||
* \brief Volk puppet for the multiple 16-bit complex dot product kernel.
|
||||
* \authors <ul>
|
||||
* <li> Carles Fernandez Prades 2016 cfernandez at cttc dot cat
|
||||
* </ul>
|
||||
*
|
||||
* Volk puppet for integrating the resampler into volk's 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_16i_rotator_dotprodxnpuppet_16ic_H
|
||||
#define INCLUDED_volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic_H
|
||||
|
||||
#include "volk_gnsssdr/volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn.h"
|
||||
#include <volk_gnsssdr/volk_gnsssdr_malloc.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr.h>
|
||||
#include <string.h>
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
static inline void volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic_generic(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
|
||||
{
|
||||
// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
float rem_carrier_phase_in_rad = 0.345;
|
||||
float phase_step_rad = 0.1;
|
||||
lv_32fc_t phase[1];
|
||||
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
lv_32fc_t phase_inc[1];
|
||||
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
unsigned int n;
|
||||
int num_a_vectors = 3;
|
||||
int16_t** in_a = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
in_a[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
memcpy((int16_t*)in_a[n], (int16_t*)in, sizeof(int16_t) * num_points);
|
||||
}
|
||||
volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn_generic(result, local_code, phase_inc[0], phase,(const int16_t**) in_a, num_a_vectors, num_points);
|
||||
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(in_a[n]);
|
||||
}
|
||||
volk_gnsssdr_free(in_a);
|
||||
}
|
||||
|
||||
#endif // Generic
|
||||
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
static inline void volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic_generic_reload(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
|
||||
{
|
||||
// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
float rem_carrier_phase_in_rad = 0.345;
|
||||
float phase_step_rad = 0.1;
|
||||
lv_32fc_t phase[1];
|
||||
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
lv_32fc_t phase_inc[1];
|
||||
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
unsigned int n;
|
||||
int num_a_vectors = 3;
|
||||
int16_t** in_a = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
in_a[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
memcpy((int16_t*)in_a[n], (int16_t*)in, sizeof(int16_t) * num_points);
|
||||
}
|
||||
volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn_generic_reload(result, local_code, phase_inc[0], phase,(const int16_t**) in_a, num_a_vectors, num_points);
|
||||
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(in_a[n]);
|
||||
}
|
||||
volk_gnsssdr_free(in_a);
|
||||
}
|
||||
|
||||
#endif // Generic
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE3
|
||||
static inline void volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic_a_sse3(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
|
||||
{
|
||||
// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
float rem_carrier_phase_in_rad = 0.345;
|
||||
float phase_step_rad = 0.1;
|
||||
lv_32fc_t phase[1];
|
||||
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
lv_32fc_t phase_inc[1];
|
||||
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
unsigned int n;
|
||||
int num_a_vectors = 3;
|
||||
int16_t** in_a = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
in_a[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
memcpy((int16_t*)in_a[n], (int16_t*)in, sizeof(int16_t) * num_points);
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn_a_sse3(result, local_code, phase_inc[0], phase, (const int16_t**) in_a, num_a_vectors, num_points);
|
||||
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(in_a[n]);
|
||||
}
|
||||
volk_gnsssdr_free(in_a);
|
||||
}
|
||||
|
||||
#endif // SSE3
|
||||
|
||||
|
||||
//#ifdef LV_HAVE_SSE3
|
||||
//static inline void volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic_a_sse3_reload(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
|
||||
//{
|
||||
//// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
//float rem_carrier_phase_in_rad = 0.345;
|
||||
//float phase_step_rad = 0.1;
|
||||
//lv_32fc_t phase[1];
|
||||
//phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
//lv_32fc_t phase_inc[1];
|
||||
//phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
//unsigned int n;
|
||||
//int num_a_vectors = 3;
|
||||
//int16_t** in_a = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
//for(n = 0; n < num_a_vectors; n++)
|
||||
//{
|
||||
//in_a[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
//memcpy((int16_t*)in_a[n], (int16_t*)in, sizeof(int16_t) * num_points);
|
||||
//}
|
||||
|
||||
//volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn_a_sse3_reload(result, local_code, phase_inc[0], phase, (const int16_t**) in_a, num_a_vectors, num_points);
|
||||
|
||||
//for(n = 0; n < num_a_vectors; n++)
|
||||
//{
|
||||
//volk_gnsssdr_free(in_a[n]);
|
||||
//}
|
||||
//volk_gnsssdr_free(in_a);
|
||||
//}
|
||||
|
||||
//#endif // SSE3
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE3
|
||||
static inline void volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic_u_sse3(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
|
||||
{
|
||||
// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
float rem_carrier_phase_in_rad = 0.345;
|
||||
float phase_step_rad = 0.1;
|
||||
lv_32fc_t phase[1];
|
||||
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
lv_32fc_t phase_inc[1];
|
||||
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
unsigned int n;
|
||||
int num_a_vectors = 3;
|
||||
int16_t** in_a = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
in_a[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
memcpy((int16_t*)in_a[n], (int16_t*)in, sizeof(int16_t) * num_points);
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn_u_sse3(result, local_code, phase_inc[0], phase, (const int16_t**) in_a, num_a_vectors, num_points);
|
||||
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(in_a[n]);
|
||||
}
|
||||
volk_gnsssdr_free(in_a);
|
||||
}
|
||||
|
||||
#endif // SSE3
|
||||
|
||||
|
||||
#ifdef LV_HAVE_AVX2
|
||||
static inline void volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic_a_avx2(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
|
||||
{
|
||||
// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
float rem_carrier_phase_in_rad = 0.345;
|
||||
float phase_step_rad = 0.1;
|
||||
lv_32fc_t phase[1];
|
||||
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
lv_32fc_t phase_inc[1];
|
||||
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
unsigned int n;
|
||||
int num_a_vectors = 3;
|
||||
int16_t** in_a = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
in_a[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
memcpy((int16_t*)in_a[n], (int16_t*)in, sizeof(int16_t) * num_points);
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn_a_avx2(result, local_code, phase_inc[0], phase, (const int16_t**) in_a, num_a_vectors, num_points);
|
||||
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(in_a[n]);
|
||||
}
|
||||
volk_gnsssdr_free(in_a);
|
||||
}
|
||||
|
||||
#endif // AVX2
|
||||
|
||||
|
||||
//#ifdef LV_HAVE_AVX2
|
||||
//static inline void volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic_a_avx2_reload(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
|
||||
//{
|
||||
//// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
//float rem_carrier_phase_in_rad = 0.345;
|
||||
//float phase_step_rad = 0.1;
|
||||
//lv_32fc_t phase[1];
|
||||
//phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
//lv_32fc_t phase_inc[1];
|
||||
//phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
//unsigned int n;
|
||||
//int num_a_vectors = 3;
|
||||
//int16_t** in_a = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
//for(n = 0; n < num_a_vectors; n++)
|
||||
//{
|
||||
//in_a[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
//memcpy((int16_t*)in_a[n], (int16_t*)in, sizeof(int16_t) * num_points);
|
||||
//}
|
||||
|
||||
//volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn_a_avx2_reload(result, local_code, phase_inc[0], phase, (const int16_t**) in_a, num_a_vectors, num_points);
|
||||
|
||||
//for(n = 0; n < num_a_vectors; n++)
|
||||
//{
|
||||
//volk_gnsssdr_free(in_a[n]);
|
||||
//}
|
||||
//volk_gnsssdr_free(in_a);
|
||||
//}
|
||||
|
||||
//#endif // AVX2
|
||||
|
||||
|
||||
#ifdef LV_HAVE_AVX2
|
||||
static inline void volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic_u_avx2(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
|
||||
{
|
||||
// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
float rem_carrier_phase_in_rad = 0.345;
|
||||
float phase_step_rad = 0.1;
|
||||
lv_32fc_t phase[1];
|
||||
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
lv_32fc_t phase_inc[1];
|
||||
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
unsigned int n;
|
||||
int num_a_vectors = 3;
|
||||
int16_t** in_a = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
in_a[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
memcpy((int16_t*)in_a[n], (int16_t*)in, sizeof(int16_t) * num_points);
|
||||
}
|
||||
|
||||
volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn_u_avx2(result, local_code, phase_inc[0], phase, (const int16_t**) in_a, num_a_vectors, num_points);
|
||||
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(in_a[n]);
|
||||
}
|
||||
volk_gnsssdr_free(in_a);
|
||||
}
|
||||
|
||||
#endif // AVX2
|
||||
|
||||
|
||||
//#ifdef LV_HAVE_AVX2
|
||||
//static inline void volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic_u_avx2_reload(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
|
||||
//{
|
||||
//// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
//float rem_carrier_phase_in_rad = 0.345;
|
||||
//float phase_step_rad = 0.1;
|
||||
//lv_32fc_t phase[1];
|
||||
//phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
//lv_32fc_t phase_inc[1];
|
||||
//phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
//unsigned int n;
|
||||
//int num_a_vectors = 3;
|
||||
//int16_t** in_a = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
//for(n = 0; n < num_a_vectors; n++)
|
||||
//{
|
||||
//in_a[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
//memcpy((int16_t*)in_a[n], (int16_t*)in, sizeof(int16_t) * num_points);
|
||||
//}
|
||||
|
||||
//volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn_a_avx2_reload(result, local_code, phase_inc[0], phase, (const int16_t**) in_a, num_a_vectors, num_points);
|
||||
|
||||
//for(n = 0; n < num_a_vectors; n++)
|
||||
//{
|
||||
//volk_gnsssdr_free(in_a[n]);
|
||||
//}
|
||||
//volk_gnsssdr_free(in_a);
|
||||
//}
|
||||
|
||||
//#endif // AVX2
|
||||
|
||||
|
||||
//#ifdef LV_HAVE_NEON
|
||||
//static inline void volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic_neon(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
|
||||
//{
|
||||
//// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
//float rem_carrier_phase_in_rad = 0.345;
|
||||
//float phase_step_rad = 0.1;
|
||||
//lv_32fc_t phase[1];
|
||||
//phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
//lv_32fc_t phase_inc[1];
|
||||
//phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
//unsigned int n;
|
||||
//int num_a_vectors = 3;
|
||||
//int16_t** in_a = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
//for(n = 0; n < num_a_vectors; n++)
|
||||
//{
|
||||
//in_a[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
//memcpy((int16_t*)in_a[n], (int16_t*)in, sizeof(int16_t) * num_points);
|
||||
//}
|
||||
|
||||
//volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn_neon(result, local_code, phase_inc[0], phase, (const int16_t**) in_a, num_a_vectors, num_points);
|
||||
|
||||
//for(n = 0; n < num_a_vectors; n++)
|
||||
//{
|
||||
//volk_gnsssdr_free(in_a[n]);
|
||||
//}
|
||||
//volk_gnsssdr_free(in_a);
|
||||
//}
|
||||
|
||||
//#endif // NEON
|
||||
|
||||
|
||||
//#ifdef LV_HAVE_NEON
|
||||
//static inline void volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic_neon_vma(lv_16sc_t* result, const lv_16sc_t* local_code, const lv_16sc_t* in, unsigned int num_points)
|
||||
//{
|
||||
//// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
//float rem_carrier_phase_in_rad = 0.345;
|
||||
//float phase_step_rad = 0.1;
|
||||
//lv_32fc_t phase[1];
|
||||
//phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
//lv_32fc_t phase_inc[1];
|
||||
//phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
//unsigned int n;
|
||||
//int num_a_vectors = 3;
|
||||
//int16_t** in_a = (int16_t**)volk_gnsssdr_malloc(sizeof(int16_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
//for(n = 0; n < num_a_vectors; n++)
|
||||
//{
|
||||
//in_a[n] = (int16_t*)volk_gnsssdr_malloc(sizeof(int16_t) * num_points, volk_gnsssdr_get_alignment());
|
||||
//memcpy((int16_t*)in_a[n], (int16_t*)in, sizeof(int16_t) * num_points);
|
||||
//}
|
||||
|
||||
//volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn_neon_vma(result, local_code, phase_inc[0], phase, (const int16_t**) in_a, num_a_vectors, num_points);
|
||||
|
||||
//for(n = 0; n < num_a_vectors; n++)
|
||||
//{
|
||||
//volk_gnsssdr_free(in_a[n]);
|
||||
//}
|
||||
//volk_gnsssdr_free(in_a);
|
||||
//}
|
||||
|
||||
//#endif // NEON
|
||||
|
||||
#endif // INCLUDED_volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic_H
|
||||
|
||||
|
||||
|
||||
|
|
@ -44,8 +44,8 @@
|
|||
#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 = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
unsigned int n;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
|
|
@ -74,8 +74,8 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_generic(lv_16sc_t* r
|
|||
#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 = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -103,8 +103,8 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_a_sse3(lv_16sc_t* re
|
|||
#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;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -133,8 +133,8 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_u_sse3(lv_16sc_t* re
|
|||
#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 = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -163,8 +163,8 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_u_sse4_1(lv_16sc_t*
|
|||
#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;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -193,8 +193,8 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_a_sse4_1(lv_16sc_t*
|
|||
#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;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -223,8 +223,8 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_u_avx(lv_16sc_t* res
|
|||
#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;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -253,8 +253,8 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_a_avx(lv_16sc_t* res
|
|||
#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 = -0.6;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
|
|||
|
|
@ -0,0 +1,279 @@
|
|||
/*!
|
||||
* \file volk_gnsssdr_32f_resamplerxnpuppet_32f.h
|
||||
* \brief VOLK_GNSSSDR puppet for the multiple 32-bit float vector resampler kernel.
|
||||
* \authors <ul>
|
||||
* <li> Cillian O'Driscoll 2017 cillian.odriscoll at gmail dot com
|
||||
* </ul>
|
||||
*
|
||||
* VOLK_GNSSSDR puppet for integrating the multiple resampler into the test system
|
||||
*
|
||||
* -------------------------------------------------------------------------
|
||||
*
|
||||
* Copyright (C) 2010-2017 (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_32f_resamplerxnpuppet_32f_H
|
||||
#define INCLUDED_volk_gnsssdr_32f_resamplerxnpuppet_32f_H
|
||||
|
||||
#include "volk_gnsssdr/volk_gnsssdr_32f_xn_resampler_32f_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_32f_resamplerxnpuppet_32f_generic(float* result, const float* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
float** result_aux = (float**)volk_gnsssdr_malloc(sizeof(float*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (float*)volk_gnsssdr_malloc(sizeof(float) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_32f_xn_resampler_32f_xn_generic(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((float*)result, (float*)result_aux[0], sizeof(float) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
|
||||
#endif /* LV_HAVE_GENERIC */
|
||||
|
||||
#ifdef LV_HAVE_SSE3
|
||||
static inline void volk_gnsssdr_32f_resamplerxnpuppet_32f_a_sse3(float* result, const float* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
float** result_aux = (float**)volk_gnsssdr_malloc(sizeof(float*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (float*)volk_gnsssdr_malloc(sizeof(float) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_32f_xn_resampler_32f_xn_a_sse3(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((float*)result, (float*)result_aux[0], sizeof(float) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
#ifdef LV_HAVE_SSE3
|
||||
static inline void volk_gnsssdr_32f_resamplerxnpuppet_32f_u_sse3(float* result, const float* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
float** result_aux = (float**)volk_gnsssdr_malloc(sizeof(float*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (float*)volk_gnsssdr_malloc(sizeof(float) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_32f_xn_resampler_32f_xn_u_sse3(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((float*)result, (float*)result_aux[0], sizeof(float) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
static inline void volk_gnsssdr_32f_resamplerxnpuppet_32f_u_sse4_1(float* result, const float* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
float** result_aux = (float**)volk_gnsssdr_malloc(sizeof(float*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (float*)volk_gnsssdr_malloc(sizeof(float) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_32f_xn_resampler_32f_xn_u_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((float*)result, (float*)result_aux[0], sizeof(float) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
static inline void volk_gnsssdr_32f_resamplerxnpuppet_32f_a_sse4_1(float* result, const float* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
float** result_aux = (float**)volk_gnsssdr_malloc(sizeof(float*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (float*)volk_gnsssdr_malloc(sizeof(float) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_32f_xn_resampler_32f_xn_a_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((float*)result, (float*)result_aux[0], sizeof(float) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
#ifdef LV_HAVE_AVX
|
||||
static inline void volk_gnsssdr_32f_resamplerxnpuppet_32f_a_avx(float* result, const float* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
float** result_aux = (float**)volk_gnsssdr_malloc(sizeof(float*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (float*)volk_gnsssdr_malloc(sizeof(float) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_32f_xn_resampler_32f_xn_a_avx(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((float*)result, (float*)result_aux[0], sizeof(float) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_AVX
|
||||
static inline void volk_gnsssdr_32f_resamplerxnpuppet_32f_u_avx(float* result, const float* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
float** result_aux = (float**)volk_gnsssdr_malloc(sizeof(float*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (float*)volk_gnsssdr_malloc(sizeof(float) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_32f_xn_resampler_32f_xn_u_avx(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((float*)result, (float*)result_aux[0], sizeof(float) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
#endif
|
||||
|
||||
#ifdef LV_HAVE_NEON
|
||||
static inline void volk_gnsssdr_32f_resamplerxnpuppet_32f_neon(float* result, const float* local_code, unsigned int num_points)
|
||||
{
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
float shifts_chips[3] = { -0.1, 0.0, 0.1 };
|
||||
|
||||
float** result_aux = (float**)volk_gnsssdr_malloc(sizeof(float*) * num_out_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
result_aux[n] = (float*)volk_gnsssdr_malloc(sizeof(float) * num_points, volk_gnsssdr_get_alignment());
|
||||
}
|
||||
|
||||
volk_gnsssdr_32f_xn_resampler_32f_xn_neon(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
|
||||
|
||||
memcpy((float*)result, (float*)result_aux[0], sizeof(float) * num_points);
|
||||
|
||||
for(n = 0; n < num_out_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(result_aux[n]);
|
||||
}
|
||||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
#endif
|
||||
|
||||
#endif // INCLUDED_volk_gnsssdr_32f_resamplerpuppet_32f_H
|
||||
|
||||
|
|
@ -0,0 +1,610 @@
|
|||
/*!
|
||||
* \file volk_gnsssdr_32f_xn_resampler_32f_xn.h
|
||||
* \brief VOLK_GNSSSDR kernel: Resamples N complex 32-bit float vectors using zero hold resample algorithm.
|
||||
* \authors <ul>
|
||||
* <li> Cillian O'Driscoll, 2017. cillian.odirscoll(at)gmail.com
|
||||
* </ul>
|
||||
*
|
||||
* VOLK_GNSSSDR kernel that resamples N 32-bit float vectors using zero hold resample algorithm.
|
||||
* It is optimized to resample 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) 2010-2017 (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_32f_xn_resampler_32f_xn
|
||||
*
|
||||
* \b Overview
|
||||
*
|
||||
* Resamples a 32-bit floating point vector , providing \p num_out_vectors outputs.
|
||||
*
|
||||
* <b>Dispatcher Prototype</b>
|
||||
* \code
|
||||
* void volk_gnsssdr_32f_xn_resampler_32f_xn(float** result, const float* local_code, float rem_code_phase_chips, float code_phase_step_chips, float* shifts_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_points)
|
||||
* \endcode
|
||||
*
|
||||
* \b Inputs
|
||||
* \li local_code: Vector to be resampled.
|
||||
* \li rem_code_phase_chips: Remnant code phase [chips].
|
||||
* \li code_phase_step_chips: Phase increment per sample [chips/sample].
|
||||
* \li shifts_chips: Vector of floats that defines the spacing (in chips) between the replicas of \p local_code
|
||||
* \li code_length_chips: Code length in chips.
|
||||
* \li num_out_vectors Number of output vectors.
|
||||
* \li num_points: The number of data values to be in the resampled vector.
|
||||
*
|
||||
* \b Outputs
|
||||
* \li result: Pointer to a vector of pointers where the results will be stored.
|
||||
*
|
||||
*/
|
||||
|
||||
#ifndef INCLUDED_volk_gnsssdr_32f_xn_resampler_32f_xn_H
|
||||
#define INCLUDED_volk_gnsssdr_32f_xn_resampler_32f_xn_H
|
||||
|
||||
#include <assert.h>
|
||||
#include <math.h>
|
||||
#include <stdlib.h> /* abs */
|
||||
#include <stdint.h> /* int64_t */
|
||||
#include <stdio.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_common.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
|
||||
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
|
||||
static inline void volk_gnsssdr_32f_xn_resampler_32f_xn_generic(float** result, const float* local_code, float rem_code_phase_chips, float code_phase_step_chips, float* shifts_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_points)
|
||||
{
|
||||
int local_code_chip_index;
|
||||
int current_correlator_tap;
|
||||
int n;
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
for (n = 0; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index = (int)floor(code_phase_step_chips * (float)n + shifts_chips[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_SSE3
|
||||
#include <pmmintrin.h>
|
||||
static inline void volk_gnsssdr_32f_xn_resampler_32f_xn_a_sse3(float** result, const float* local_code, float rem_code_phase_chips, float code_phase_step_chips, float* shifts_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_points)
|
||||
{
|
||||
float** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
const __m128 ones = _mm_set1_ps(1.0f);
|
||||
const __m128 fours = _mm_set1_ps(4.0f);
|
||||
const __m128 rem_code_phase_chips_reg = _mm_set_ps1(rem_code_phase_chips);
|
||||
const __m128 code_phase_step_chips_reg = _mm_set_ps1(code_phase_step_chips);
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) int local_code_chip_index[4];
|
||||
int local_code_chip_index_;
|
||||
|
||||
const __m128i zeros = _mm_setzero_si128();
|
||||
const __m128 code_length_chips_reg_f = _mm_set_ps1((float)code_length_chips);
|
||||
const __m128i code_length_chips_reg_i = _mm_set1_epi32((int)code_length_chips);
|
||||
__m128i local_code_chip_index_reg, aux_i, negatives, i;
|
||||
__m128 aux, aux2, shifts_chips_reg, fi, igx, j, c, cTrunc, base;
|
||||
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
shifts_chips_reg = _mm_set_ps1((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
__m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
|
||||
for(n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
i = _mm_cvttps_epi32(aux);
|
||||
fi = _mm_cvtepi32_ps(i);
|
||||
igx = _mm_cmpgt_ps(fi, aux);
|
||||
j = _mm_and_ps(igx, ones);
|
||||
aux = _mm_sub_ps(fi, j);
|
||||
// fmod
|
||||
c = _mm_div_ps(aux, code_length_chips_reg_f);
|
||||
i = _mm_cvttps_epi32(c);
|
||||
cTrunc = _mm_cvtepi32_ps(i);
|
||||
base = _mm_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||
local_code_chip_index_reg = _mm_cvtps_epi32(_mm_sub_ps(aux, base));
|
||||
|
||||
negatives = _mm_cmplt_epi32(local_code_chip_index_reg, zeros);
|
||||
aux_i = _mm_and_si128(code_length_chips_reg_i, negatives);
|
||||
local_code_chip_index_reg = _mm_add_epi32(local_code_chip_index_reg, aux_i);
|
||||
_mm_store_si128((__m128i*)local_code_chip_index, local_code_chip_index_reg);
|
||||
for(k = 0; k < 4; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm_add_ps(indexn, fours);
|
||||
}
|
||||
for(n = quarterPoints * 4; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
|
||||
//Take into account that in multitap correlators, the shifts can be negative!
|
||||
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1) ;
|
||||
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE3
|
||||
#include <pmmintrin.h>
|
||||
static inline void volk_gnsssdr_32f_xn_resampler_32f_xn_u_sse3(float** result, const float* local_code, float rem_code_phase_chips, float code_phase_step_chips, float* shifts_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_points)
|
||||
{
|
||||
float** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
const __m128 ones = _mm_set1_ps(1.0f);
|
||||
const __m128 fours = _mm_set1_ps(4.0f);
|
||||
const __m128 rem_code_phase_chips_reg = _mm_set_ps1(rem_code_phase_chips);
|
||||
const __m128 code_phase_step_chips_reg = _mm_set_ps1(code_phase_step_chips);
|
||||
|
||||
__VOLK_ATTR_ALIGNED(16) int local_code_chip_index[4];
|
||||
int local_code_chip_index_;
|
||||
|
||||
const __m128i zeros = _mm_setzero_si128();
|
||||
const __m128 code_length_chips_reg_f = _mm_set_ps1((float)code_length_chips);
|
||||
const __m128i code_length_chips_reg_i = _mm_set1_epi32((int)code_length_chips);
|
||||
__m128i local_code_chip_index_reg, aux_i, negatives, i;
|
||||
__m128 aux, aux2, shifts_chips_reg, fi, igx, j, c, cTrunc, base;
|
||||
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
shifts_chips_reg = _mm_set_ps1((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
__m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
|
||||
for(n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
i = _mm_cvttps_epi32(aux);
|
||||
fi = _mm_cvtepi32_ps(i);
|
||||
igx = _mm_cmpgt_ps(fi, aux);
|
||||
j = _mm_and_ps(igx, ones);
|
||||
aux = _mm_sub_ps(fi, j);
|
||||
// fmod
|
||||
c = _mm_div_ps(aux, code_length_chips_reg_f);
|
||||
i = _mm_cvttps_epi32(c);
|
||||
cTrunc = _mm_cvtepi32_ps(i);
|
||||
base = _mm_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||
local_code_chip_index_reg = _mm_cvtps_epi32(_mm_sub_ps(aux, base));
|
||||
|
||||
negatives = _mm_cmplt_epi32(local_code_chip_index_reg, zeros);
|
||||
aux_i = _mm_and_si128(code_length_chips_reg_i, negatives);
|
||||
local_code_chip_index_reg = _mm_add_epi32(local_code_chip_index_reg, aux_i);
|
||||
_mm_store_si128((__m128i*)local_code_chip_index, local_code_chip_index_reg);
|
||||
for(k = 0; k < 4; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm_add_ps(indexn, fours);
|
||||
}
|
||||
for(n = quarterPoints * 4; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
|
||||
//Take into account that in multitap correlators, the shifts can be negative!
|
||||
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1) ;
|
||||
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||
}
|
||||
}
|
||||
}
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
#include <smmintrin.h>
|
||||
static inline void volk_gnsssdr_32f_xn_resampler_32f_xn_a_sse4_1(float** result, const float* local_code, float rem_code_phase_chips, float code_phase_step_chips, float* shifts_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_points)
|
||||
{
|
||||
float** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
const __m128 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 (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
shifts_chips_reg = _mm_set_ps1((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
__m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
|
||||
for(n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
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(k = 0; k < 4; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm_add_ps(indexn, fours);
|
||||
}
|
||||
for(n = quarterPoints * 4; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
|
||||
//Take into account that in multitap correlators, the shifts can be negative!
|
||||
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1) ;
|
||||
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE4_1
|
||||
#include <smmintrin.h>
|
||||
static inline void volk_gnsssdr_32f_xn_resampler_32f_xn_u_sse4_1(float** result, const float* local_code, float rem_code_phase_chips, float code_phase_step_chips, float* shifts_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_points)
|
||||
{
|
||||
float** _result = result;
|
||||
const unsigned int quarterPoints = num_points / 4;
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
const __m128 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 (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
shifts_chips_reg = _mm_set_ps1((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
__m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
|
||||
for(n = 0; n < quarterPoints; n++)
|
||||
{
|
||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm_add_ps(aux, aux2);
|
||||
// floor
|
||||
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(k = 0; k < 4; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm_add_ps(indexn, fours);
|
||||
}
|
||||
for(n = quarterPoints * 4; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
|
||||
//Take into account that in multitap correlators, the shifts can be negative!
|
||||
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1) ;
|
||||
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_AVX
|
||||
#include <immintrin.h>
|
||||
static inline void volk_gnsssdr_32f_xn_resampler_32f_xn_a_avx(float** result, const float* local_code, float rem_code_phase_chips, float code_phase_step_chips, float* shifts_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_points)
|
||||
{
|
||||
float** _result = result;
|
||||
const unsigned int avx_iters = num_points / 8;
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
const __m256 eights = _mm256_set1_ps(8.0f);
|
||||
const __m256 rem_code_phase_chips_reg = _mm256_set1_ps(rem_code_phase_chips);
|
||||
const __m256 code_phase_step_chips_reg = _mm256_set1_ps(code_phase_step_chips);
|
||||
|
||||
__VOLK_ATTR_ALIGNED(32) int local_code_chip_index[8];
|
||||
int local_code_chip_index_;
|
||||
|
||||
const __m256 zeros = _mm256_setzero_ps();
|
||||
const __m256 code_length_chips_reg_f = _mm256_set1_ps((float)code_length_chips);
|
||||
const __m256 n0 = _mm256_set_ps(7.0f, 6.0f, 5.0f, 4.0f, 3.0f, 2.0f, 1.0f, 0.0f);
|
||||
|
||||
__m256i local_code_chip_index_reg, i;
|
||||
__m256 aux, aux2, aux3, shifts_chips_reg, c, cTrunc, base, negatives, indexn;
|
||||
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
shifts_chips_reg = _mm256_set1_ps((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = _mm256_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
indexn = n0;
|
||||
for(n = 0; n < avx_iters; n++)
|
||||
{
|
||||
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&_result[current_correlator_tap][8 * n + 7], 1, 0);
|
||||
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&local_code_chip_index[8], 1, 3);
|
||||
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm256_add_ps(aux, aux2);
|
||||
// floor
|
||||
aux = _mm256_floor_ps(aux);
|
||||
|
||||
// fmod
|
||||
c = _mm256_div_ps(aux, code_length_chips_reg_f);
|
||||
i = _mm256_cvttps_epi32(c);
|
||||
cTrunc = _mm256_cvtepi32_ps(i);
|
||||
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||
local_code_chip_index_reg = _mm256_cvttps_epi32(_mm256_sub_ps(aux, base));
|
||||
|
||||
// no negatives
|
||||
c = _mm256_cvtepi32_ps(local_code_chip_index_reg);
|
||||
negatives = _mm256_cmp_ps(c, zeros, 0x01 );
|
||||
aux3 = _mm256_and_ps(code_length_chips_reg_f, negatives);
|
||||
aux = _mm256_add_ps(c, aux3);
|
||||
local_code_chip_index_reg = _mm256_cvttps_epi32(aux);
|
||||
|
||||
_mm256_store_si256((__m256i*)local_code_chip_index, local_code_chip_index_reg);
|
||||
for(k = 0; k < 8; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 8 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm256_add_ps(indexn, eights);
|
||||
}
|
||||
}
|
||||
_mm256_zeroupper();
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
for(n = avx_iters * 8; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
|
||||
//Take into account that in multitap correlators, the shifts can be negative!
|
||||
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1) ;
|
||||
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_AVX
|
||||
#include <immintrin.h>
|
||||
static inline void volk_gnsssdr_32f_xn_resampler_32f_xn_u_avx(float** result, const float* local_code, float rem_code_phase_chips, float code_phase_step_chips, float* shifts_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_points)
|
||||
{
|
||||
float** _result = result;
|
||||
const unsigned int avx_iters = num_points / 8;
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
const __m256 eights = _mm256_set1_ps(8.0f);
|
||||
const __m256 rem_code_phase_chips_reg = _mm256_set1_ps(rem_code_phase_chips);
|
||||
const __m256 code_phase_step_chips_reg = _mm256_set1_ps(code_phase_step_chips);
|
||||
|
||||
__VOLK_ATTR_ALIGNED(32) int local_code_chip_index[8];
|
||||
int local_code_chip_index_;
|
||||
|
||||
const __m256 zeros = _mm256_setzero_ps();
|
||||
const __m256 code_length_chips_reg_f = _mm256_set1_ps((float)code_length_chips);
|
||||
const __m256 n0 = _mm256_set_ps(7.0f, 6.0f, 5.0f, 4.0f, 3.0f, 2.0f, 1.0f, 0.0f);
|
||||
|
||||
__m256i local_code_chip_index_reg, i;
|
||||
__m256 aux, aux2, aux3, shifts_chips_reg, c, cTrunc, base, negatives, indexn;
|
||||
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
shifts_chips_reg = _mm256_set1_ps((float)shifts_chips[current_correlator_tap]);
|
||||
aux2 = _mm256_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||
indexn = n0;
|
||||
for(n = 0; n < avx_iters; n++)
|
||||
{
|
||||
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&_result[current_correlator_tap][8 * n + 7], 1, 0);
|
||||
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&local_code_chip_index[8], 1, 3);
|
||||
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
|
||||
aux = _mm256_add_ps(aux, aux2);
|
||||
// floor
|
||||
aux = _mm256_floor_ps(aux);
|
||||
|
||||
// fmod
|
||||
c = _mm256_div_ps(aux, code_length_chips_reg_f);
|
||||
i = _mm256_cvttps_epi32(c);
|
||||
cTrunc = _mm256_cvtepi32_ps(i);
|
||||
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||
local_code_chip_index_reg = _mm256_cvttps_epi32(_mm256_sub_ps(aux, base));
|
||||
|
||||
// no negatives
|
||||
c = _mm256_cvtepi32_ps(local_code_chip_index_reg);
|
||||
negatives = _mm256_cmp_ps(c, zeros, 0x01 );
|
||||
aux3 = _mm256_and_ps(code_length_chips_reg_f, negatives);
|
||||
aux = _mm256_add_ps(c, aux3);
|
||||
local_code_chip_index_reg = _mm256_cvttps_epi32(aux);
|
||||
|
||||
_mm256_store_si256((__m256i*)local_code_chip_index, local_code_chip_index_reg);
|
||||
for(k = 0; k < 8; ++k)
|
||||
{
|
||||
_result[current_correlator_tap][n * 8 + k] = local_code[local_code_chip_index[k]];
|
||||
}
|
||||
indexn = _mm256_add_ps(indexn, eights);
|
||||
}
|
||||
}
|
||||
_mm256_zeroupper();
|
||||
for (current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
||||
{
|
||||
for(n = avx_iters * 8; n < num_points; n++)
|
||||
{
|
||||
// resample code for current tap
|
||||
local_code_chip_index_ = (int)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
|
||||
//Take into account that in multitap correlators, the shifts can be negative!
|
||||
if (local_code_chip_index_ < 0) local_code_chip_index_ += (int)code_length_chips * (abs(local_code_chip_index_) / code_length_chips + 1) ;
|
||||
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
|
||||
#ifdef LV_HAVE_NEON
|
||||
#include <arm_neon.h>
|
||||
|
||||
static inline void volk_gnsssdr_32f_xn_resampler_32f_xn_neon(float** result, const float* local_code, float rem_code_phase_chips, float code_phase_step_chips, float* shifts_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_points)
|
||||
{
|
||||
float** _result = result;
|
||||
const unsigned int neon_iters = num_points / 4;
|
||||
int current_correlator_tap;
|
||||
unsigned int n;
|
||||
unsigned int k;
|
||||
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 (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(n = 0; n < neon_iters; n++)
|
||||
{
|
||||
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&_result[current_correlator_tap][4 * n + 3], 1, 0);
|
||||
__VOLK_GNSSSDR_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(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(n = neon_iters * 4; n < num_points; n++)
|
||||
{
|
||||
__VOLK_GNSSSDR_PREFETCH_LOCALITY(&_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_32f_xn_resampler_32f_xn_H*/
|
||||
|
||||
|
||||
|
|
@ -0,0 +1,320 @@
|
|||
/*!
|
||||
* \file volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn.h
|
||||
* \brief VOLK_GNSSSDR kernel: multiplies N complex (32-bit float per component) vectors
|
||||
* by a common vector, phase rotated and accumulates the results in N float complex outputs.
|
||||
* \authors <ul>
|
||||
* <li> Cillian O'Driscoll 2016. cillian.odriscoll(at)gmail.com
|
||||
* </ul>
|
||||
*
|
||||
* VOLK_GNSSSDR kernel that multiplies N 32 bits complex vectors by a common vector, which is
|
||||
* phase-rotated by phase offset and phase increment, and accumulates the results
|
||||
* in N 32 bits float complex outputs.
|
||||
* It is optimized to perform the N tap correlation process in GNSS receivers.
|
||||
*
|
||||
* -------------------------------------------------------------------------
|
||||
*
|
||||
* Copyright (C) 2010-2016 (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_32fc_32f_rotator_dot_prod_32fc_xn
|
||||
*
|
||||
* \b Overview
|
||||
*
|
||||
* Rotates and multiplies the reference complex vector with an arbitrary number of other real vectors,
|
||||
* accumulates the results and stores them in the output vector.
|
||||
* The rotation is done at a fixed rate per sample, from an initial \p phase offset.
|
||||
* This function can be used for Doppler wipe-off and multiple correlator.
|
||||
*
|
||||
* <b>Dispatcher Prototype</b>
|
||||
* \code
|
||||
* void volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn(lv_32fc_t* result, const lv_32fc_t* in_common, const lv_32fc_t phase_inc, lv_32fc_t* phase, const float** in_a, int num_a_vectors, unsigned int num_points);
|
||||
* \endcode
|
||||
*
|
||||
* \b Inputs
|
||||
* \li in_common: Pointer to one of the vectors to be rotated, multiplied and accumulated (reference vector).
|
||||
* \li phase_inc: Phase increment = lv_cmake(cos(phase_step_rad), sin(phase_step_rad))
|
||||
* \li phase: Initial phase = lv_cmake(cos(initial_phase_rad), sin(initial_phase_rad))
|
||||
* \li in_a: Pointer to an array of pointers to multiple vectors to be multiplied and accumulated.
|
||||
* \li num_a_vectors: Number of vectors to be multiplied by the reference vector and accumulated.
|
||||
* \li num_points: Number of complex values to be multiplied together, accumulated and stored into \p result.
|
||||
*
|
||||
* \b Outputs
|
||||
* \li phase: Final phase.
|
||||
* \li result: Vector of \p num_a_vectors components with the multiple vectors of \p in_a rotated, multiplied by \p in_common and accumulated.
|
||||
*
|
||||
*/
|
||||
|
||||
#ifndef INCLUDED_volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn_H
|
||||
#define INCLUDED_volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn_H
|
||||
|
||||
|
||||
#include <volk_gnsssdr/volk_gnsssdr.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_malloc.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
|
||||
#include <volk_gnsssdr/saturation_arithmetic.h>
|
||||
#include <math.h>
|
||||
#include <stdio.h>
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
|
||||
static inline void volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn_generic(lv_32fc_t* result, const lv_32fc_t* in_common, const lv_32fc_t phase_inc, lv_32fc_t* phase, const float** in_a, int num_a_vectors, unsigned int num_points)
|
||||
{
|
||||
lv_32fc_t tmp32_1, tmp32_2;
|
||||
int n_vec;
|
||||
unsigned int n;
|
||||
for (n_vec = 0; n_vec < num_a_vectors; n_vec++)
|
||||
{
|
||||
result[n_vec] = lv_cmake(0,0);
|
||||
}
|
||||
for (n = 0; n < num_points; n++)
|
||||
{
|
||||
tmp32_1 = *in_common++ * (*phase);//if(n<10 || n >= 8108) printf("generic phase %i: %f,%f\n", n,lv_creal(*phase),lv_cimag(*phase));
|
||||
|
||||
// Regenerate phase
|
||||
if (n % 256 == 0)
|
||||
{
|
||||
//printf("Phase before regeneration %i: %f,%f Modulus: %f\n", n,lv_creal(*phase),lv_cimag(*phase), cabsf(*phase));
|
||||
#ifdef __cplusplus
|
||||
(*phase) /= std::abs((*phase));
|
||||
#else
|
||||
(*phase) /= hypotf(lv_creal(*phase), lv_cimag(*phase));
|
||||
#endif
|
||||
//printf("Phase after regeneration %i: %f,%f Modulus: %f\n", n,lv_creal(*phase),lv_cimag(*phase), cabsf(*phase));
|
||||
}
|
||||
|
||||
(*phase) *= phase_inc;
|
||||
for (n_vec = 0; n_vec < num_a_vectors; n_vec++)
|
||||
{
|
||||
tmp32_2 = tmp32_1 * in_a[n_vec][n];
|
||||
result[n_vec] += tmp32_2;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif /*LV_HAVE_GENERIC*/
|
||||
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
|
||||
static inline void volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn_generic_reload(lv_32fc_t* result, const lv_32fc_t* in_common, const lv_32fc_t phase_inc, lv_32fc_t* phase, const float** in_a, int num_a_vectors, unsigned int num_points)
|
||||
{
|
||||
lv_32fc_t tmp32_1, tmp32_2;
|
||||
const unsigned int ROTATOR_RELOAD = 256;
|
||||
int n_vec;
|
||||
unsigned int n;
|
||||
unsigned int j;
|
||||
for (n_vec = 0; n_vec < num_a_vectors; n_vec++)
|
||||
{
|
||||
result[n_vec] = lv_cmake(0,0);
|
||||
}
|
||||
|
||||
for (n = 0; n < num_points / ROTATOR_RELOAD; n++)
|
||||
{
|
||||
for (j = 0; j < ROTATOR_RELOAD; j++)
|
||||
{
|
||||
tmp32_1 = *in_common++ * (*phase);
|
||||
(*phase) *= phase_inc;
|
||||
for (n_vec = 0; n_vec < num_a_vectors; n_vec++)
|
||||
{
|
||||
tmp32_2 = tmp32_1 * in_a[n_vec][n * ROTATOR_RELOAD + j];
|
||||
result[n_vec] += tmp32_2;
|
||||
}
|
||||
}
|
||||
/* Regenerate phase */
|
||||
#ifdef __cplusplus
|
||||
(*phase) /= std::abs((*phase));
|
||||
#else
|
||||
//(*phase) /= cabsf((*phase));
|
||||
(*phase) /= hypotf(lv_creal(*phase), lv_cimag(*phase));
|
||||
#endif
|
||||
}
|
||||
|
||||
for (j = 0; j < num_points % ROTATOR_RELOAD; j++)
|
||||
{
|
||||
tmp32_1 = *in_common++ * (*phase);
|
||||
(*phase) *= phase_inc;
|
||||
for (n_vec = 0; n_vec < num_a_vectors; n_vec++)
|
||||
{
|
||||
tmp32_2 = tmp32_1 * in_a[n_vec][(num_points / ROTATOR_RELOAD) * ROTATOR_RELOAD + j];
|
||||
result[n_vec] += tmp32_2;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#endif /*LV_HAVE_GENERIC*/
|
||||
|
||||
#ifdef LV_HAVE_AVX
|
||||
#include <immintrin.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr_avx_intrinsics.h>
|
||||
static inline void volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn_u_avx(lv_32fc_t* result, const lv_32fc_t* in_common, const lv_32fc_t phase_inc, lv_32fc_t* phase, const float** in_a, int num_a_vectors, unsigned int num_points)
|
||||
{
|
||||
unsigned int number = 0;
|
||||
unsigned int vec_ind = 0;
|
||||
unsigned int i = 0;
|
||||
const unsigned int sixteenthPoints = num_points / 16;
|
||||
|
||||
const float* aPtr = (float*)in_common;
|
||||
const float* bPtr[ num_a_vectors];
|
||||
for( vec_ind = 0; vec_ind < num_a_vectors; ++vec_ind ){
|
||||
bPtr[vec_ind] = in_a[vec_ind];
|
||||
}
|
||||
|
||||
lv_32fc_t _phase = (*phase);
|
||||
lv_32fc_t wo;
|
||||
|
||||
__m256 a0Val, a1Val, a2Val, a3Val;
|
||||
__m256 b0Val[num_a_vectors], b1Val[num_a_vectors], b2Val[num_a_vectors], b3Val[num_a_vectors];
|
||||
__m256 x0Val[num_a_vectors], x1Val[num_a_vectors], x0loVal[num_a_vectors], x0hiVal[num_a_vectors], x1loVal[num_a_vectors], x1hiVal[num_a_vectors];
|
||||
__m256 c0Val[num_a_vectors], c1Val[num_a_vectors], c2Val[num_a_vectors], c3Val[num_a_vectors];
|
||||
|
||||
__m256 dotProdVal0[num_a_vectors];
|
||||
__m256 dotProdVal1[num_a_vectors];
|
||||
__m256 dotProdVal2[num_a_vectors];
|
||||
__m256 dotProdVal3[num_a_vectors];
|
||||
|
||||
for( vec_ind = 0; vec_ind < num_a_vectors; vec_ind++ ){
|
||||
dotProdVal0[vec_ind] = _mm256_setzero_ps();
|
||||
dotProdVal1[vec_ind] = _mm256_setzero_ps();
|
||||
dotProdVal2[vec_ind] = _mm256_setzero_ps();
|
||||
dotProdVal3[vec_ind] = _mm256_setzero_ps();
|
||||
}
|
||||
|
||||
// Set up the complex rotator
|
||||
__m256 z0, z1, z2, z3;
|
||||
__attribute__((aligned(32))) lv_32fc_t phase_vec[16];
|
||||
for( vec_ind = 0; vec_ind < 16; ++vec_ind ){
|
||||
phase_vec[vec_ind] = _phase;
|
||||
_phase *= phase_inc;
|
||||
}
|
||||
|
||||
z0 = _mm256_load_ps( (float *)phase_vec );
|
||||
z1 = _mm256_load_ps( (float *)(phase_vec + 4) );
|
||||
z2 = _mm256_load_ps( (float *)(phase_vec + 8) );
|
||||
z3 = _mm256_load_ps( (float *)(phase_vec + 12) );
|
||||
|
||||
lv_32fc_t dz = phase_inc; dz *= dz; dz *= dz; dz *= dz; dz *= dz; // dz = phase_inc^16;
|
||||
|
||||
for( vec_ind = 0; vec_ind < 4; ++vec_ind ){
|
||||
phase_vec[vec_ind] = dz;
|
||||
}
|
||||
|
||||
__m256 dz_reg = _mm256_load_ps( (float *)phase_vec );
|
||||
dz_reg = _mm256_complexnormalise_ps( dz_reg );
|
||||
|
||||
for(;number < sixteenthPoints; number++){
|
||||
|
||||
a0Val = _mm256_loadu_ps(aPtr);
|
||||
a1Val = _mm256_loadu_ps(aPtr+8);
|
||||
a2Val = _mm256_loadu_ps(aPtr+16);
|
||||
a3Val = _mm256_loadu_ps(aPtr+24);
|
||||
|
||||
a0Val = _mm256_complexmul_ps( a0Val, z0 );
|
||||
a1Val = _mm256_complexmul_ps( a1Val, z1 );
|
||||
a2Val = _mm256_complexmul_ps( a2Val, z2 );
|
||||
a3Val = _mm256_complexmul_ps( a3Val, z3 );
|
||||
|
||||
z0 = _mm256_complexmul_ps( z0, dz_reg );
|
||||
z1 = _mm256_complexmul_ps( z1, dz_reg );
|
||||
z2 = _mm256_complexmul_ps( z2, dz_reg );
|
||||
z3 = _mm256_complexmul_ps( z3, dz_reg );
|
||||
|
||||
|
||||
for( vec_ind = 0; vec_ind < num_a_vectors; ++vec_ind ){
|
||||
x0Val[vec_ind] = _mm256_loadu_ps(bPtr[vec_ind]); // t0|t1|t2|t3|t4|t5|t6|t7
|
||||
x1Val[vec_ind] = _mm256_loadu_ps(bPtr[vec_ind]+8);
|
||||
x0loVal[vec_ind] = _mm256_unpacklo_ps(x0Val[vec_ind], x0Val[vec_ind]); // t0|t0|t1|t1|t4|t4|t5|t5
|
||||
x0hiVal[vec_ind] = _mm256_unpackhi_ps(x0Val[vec_ind], x0Val[vec_ind]); // t2|t2|t3|t3|t6|t6|t7|t7
|
||||
x1loVal[vec_ind] = _mm256_unpacklo_ps(x1Val[vec_ind], x1Val[vec_ind]);
|
||||
x1hiVal[vec_ind] = _mm256_unpackhi_ps(x1Val[vec_ind], x1Val[vec_ind]);
|
||||
|
||||
// TODO: it may be possible to rearrange swizzling to better pipeline data
|
||||
b0Val[vec_ind] = _mm256_permute2f128_ps(x0loVal[vec_ind], x0hiVal[vec_ind], 0x20); // t0|t0|t1|t1|t2|t2|t3|t3
|
||||
b1Val[vec_ind] = _mm256_permute2f128_ps(x0loVal[vec_ind], x0hiVal[vec_ind], 0x31); // t4|t4|t5|t5|t6|t6|t7|t7
|
||||
b2Val[vec_ind] = _mm256_permute2f128_ps(x1loVal[vec_ind], x1hiVal[vec_ind], 0x20);
|
||||
b3Val[vec_ind] = _mm256_permute2f128_ps(x1loVal[vec_ind], x1hiVal[vec_ind], 0x31);
|
||||
|
||||
c0Val[vec_ind] = _mm256_mul_ps(a0Val, b0Val[vec_ind]);
|
||||
c1Val[vec_ind] = _mm256_mul_ps(a1Val, b1Val[vec_ind]);
|
||||
c2Val[vec_ind] = _mm256_mul_ps(a2Val, b2Val[vec_ind]);
|
||||
c3Val[vec_ind] = _mm256_mul_ps(a3Val, b3Val[vec_ind]);
|
||||
|
||||
dotProdVal0[vec_ind] = _mm256_add_ps(c0Val[vec_ind], dotProdVal0[vec_ind]);
|
||||
dotProdVal1[vec_ind] = _mm256_add_ps(c1Val[vec_ind], dotProdVal1[vec_ind]);
|
||||
dotProdVal2[vec_ind] = _mm256_add_ps(c2Val[vec_ind], dotProdVal2[vec_ind]);
|
||||
dotProdVal3[vec_ind] = _mm256_add_ps(c3Val[vec_ind], dotProdVal3[vec_ind]);
|
||||
|
||||
bPtr[vec_ind] += 16;
|
||||
}
|
||||
|
||||
// Force the rotators back onto the unit circle
|
||||
if ((number % 64) == 0)
|
||||
{
|
||||
z0 = _mm256_complexnormalise_ps( z0 );
|
||||
z1 = _mm256_complexnormalise_ps( z1 );
|
||||
z2 = _mm256_complexnormalise_ps( z2 );
|
||||
z3 = _mm256_complexnormalise_ps( z3 );
|
||||
}
|
||||
|
||||
aPtr += 32;
|
||||
}
|
||||
__VOLK_ATTR_ALIGNED(32) lv_32fc_t dotProductVector[4];
|
||||
|
||||
for( vec_ind = 0; vec_ind < num_a_vectors; ++vec_ind ){
|
||||
dotProdVal0[vec_ind] = _mm256_add_ps(dotProdVal0[vec_ind], dotProdVal1[vec_ind]);
|
||||
dotProdVal0[vec_ind] = _mm256_add_ps(dotProdVal0[vec_ind], dotProdVal2[vec_ind]);
|
||||
dotProdVal0[vec_ind] = _mm256_add_ps(dotProdVal0[vec_ind], dotProdVal3[vec_ind]);
|
||||
|
||||
_mm256_store_ps((float *)dotProductVector,dotProdVal0[vec_ind]); // Store the results back into the dot product vector
|
||||
|
||||
result[ vec_ind ] = lv_cmake( 0, 0 );
|
||||
for( i = 0; i < 4; ++i ){
|
||||
result[vec_ind] += dotProductVector[i];
|
||||
}
|
||||
}
|
||||
|
||||
z0 = _mm256_complexnormalise_ps( z0 );
|
||||
_mm256_store_ps((float*)phase_vec, z0);
|
||||
_phase = phase_vec[0];
|
||||
_mm256_zeroupper();
|
||||
|
||||
|
||||
number = sixteenthPoints*16;
|
||||
for(;number < num_points; number++){
|
||||
wo = (*aPtr++)*_phase;
|
||||
_phase *= phase_inc;
|
||||
|
||||
for( vec_ind = 0; vec_ind < num_a_vectors; ++vec_ind ){
|
||||
result[vec_ind] += wo * in_a[vec_ind][number];
|
||||
}
|
||||
}
|
||||
|
||||
*phase = _phase;
|
||||
|
||||
}
|
||||
|
||||
#endif /* LV_HAVE_AVX */
|
||||
|
||||
#endif /* INCLUDED_volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn_H */
|
||||
|
||||
|
||||
|
|
@ -0,0 +1,132 @@
|
|||
/*!
|
||||
* \file volk_gnsssdr_32fc_32f_rotator_dotprodxnpuppet_32fc.h
|
||||
* \brief Volk puppet for the multiple 16-bit complex dot product kernel.
|
||||
* \authors <ul>
|
||||
* <li> Carles Fernandez Prades 2016 cfernandez at cttc dot cat
|
||||
* </ul>
|
||||
*
|
||||
* Volk puppet for integrating the resampler into volk's 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_32fc_32f_rotator_dotprodxnpuppet_32fc_H
|
||||
#define INCLUDED_volk_gnsssdr_32fc_32f_rotator_dotprodxnpuppet_32fc_H
|
||||
|
||||
#include "volk_gnsssdr/volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn.h"
|
||||
#include <volk_gnsssdr/volk_gnsssdr_malloc.h>
|
||||
#include <volk_gnsssdr/volk_gnsssdr.h>
|
||||
#include <string.h>
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
|
||||
static inline void volk_gnsssdr_32fc_32f_rotator_dotprodxnpuppet_32fc_generic(lv_32fc_t* result, const lv_32fc_t* local_code, const float* in, unsigned int num_points)
|
||||
{
|
||||
// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
float rem_carrier_phase_in_rad = 0.25;
|
||||
float phase_step_rad = 0.1;
|
||||
lv_32fc_t phase[1];
|
||||
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
lv_32fc_t phase_inc[1];
|
||||
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
unsigned int n;
|
||||
int num_a_vectors = 3;
|
||||
float ** in_a = (float **)volk_gnsssdr_malloc(sizeof(float *) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
in_a[n] = (float *)volk_gnsssdr_malloc(sizeof(float ) * num_points, volk_gnsssdr_get_alignment());
|
||||
memcpy((float*)in_a[n], (float*)in, sizeof(float) * num_points);
|
||||
}
|
||||
volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn_generic(result, local_code, phase_inc[0], phase, (const float**) in_a, num_a_vectors, num_points);
|
||||
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(in_a[n]);
|
||||
}
|
||||
volk_gnsssdr_free(in_a);
|
||||
}
|
||||
#endif // Generic
|
||||
|
||||
|
||||
#ifdef LV_HAVE_GENERIC
|
||||
static inline void volk_gnsssdr_32fc_32f_rotator_dotprodxnpuppet_32fc_generic_reload(lv_32fc_t* result, const lv_32fc_t* local_code, const float* in, unsigned int num_points)
|
||||
{
|
||||
// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
float rem_carrier_phase_in_rad = 0.25;
|
||||
float phase_step_rad = 0.1;
|
||||
lv_32fc_t phase[1];
|
||||
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
lv_32fc_t phase_inc[1];
|
||||
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
unsigned int n;
|
||||
int num_a_vectors = 3;
|
||||
float ** in_a = (float **)volk_gnsssdr_malloc(sizeof(float *) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
in_a[n] = (float *)volk_gnsssdr_malloc(sizeof(float ) * num_points, volk_gnsssdr_get_alignment());
|
||||
memcpy((float*)in_a[n], (float*)in, sizeof(float) * num_points);
|
||||
}
|
||||
volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn_generic_reload(result, local_code, phase_inc[0], phase, (const float**) in_a, num_a_vectors, num_points);
|
||||
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(in_a[n]);
|
||||
}
|
||||
volk_gnsssdr_free(in_a);
|
||||
}
|
||||
|
||||
#endif // Generic
|
||||
|
||||
#ifdef LV_HAVE_AVX
|
||||
static inline void volk_gnsssdr_32fc_32f_rotator_dotprodxnpuppet_32fc_u_avx(lv_32fc_t* result, const lv_32fc_t* local_code, const float* in, unsigned int num_points)
|
||||
{
|
||||
// phases must be normalized. Phase rotator expects a complex exponential input!
|
||||
float rem_carrier_phase_in_rad = 0.25;
|
||||
float phase_step_rad = 0.1;
|
||||
lv_32fc_t phase[1];
|
||||
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), sin(rem_carrier_phase_in_rad));
|
||||
lv_32fc_t phase_inc[1];
|
||||
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
|
||||
unsigned int n;
|
||||
int num_a_vectors = 3;
|
||||
float ** in_a = (float **)volk_gnsssdr_malloc(sizeof(float *) * num_a_vectors, volk_gnsssdr_get_alignment());
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
in_a[n] = (float *)volk_gnsssdr_malloc(sizeof(float ) * num_points, volk_gnsssdr_get_alignment());
|
||||
memcpy((float*)in_a[n], (float*)in, sizeof(float) * num_points);
|
||||
}
|
||||
volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn_u_avx(result, local_code, phase_inc[0], phase, (const float**) in_a, num_a_vectors, num_points);
|
||||
|
||||
for(n = 0; n < num_a_vectors; n++)
|
||||
{
|
||||
volk_gnsssdr_free(in_a[n]);
|
||||
}
|
||||
volk_gnsssdr_free(in_a);
|
||||
}
|
||||
|
||||
#endif // AVX
|
||||
|
||||
#endif // INCLUDED_volk_gnsssdr_32fc_32f_rotator_dotprodxnpuppet_32fc_H
|
||||
|
||||
|
|
@ -46,8 +46,8 @@
|
|||
#ifdef LV_HAVE_GENERIC
|
||||
static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_generic(lv_32fc_t* result, const lv_32fc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 1023;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -70,14 +70,15 @@ static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_generic(lv_32fc_t* r
|
|||
volk_gnsssdr_free(result_aux);
|
||||
}
|
||||
|
||||
|
||||
#endif /* LV_HAVE_GENERIC */
|
||||
|
||||
|
||||
#ifdef LV_HAVE_SSE3
|
||||
static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_a_sse3(lv_32fc_t* result, const lv_32fc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 1023;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -105,8 +106,8 @@ static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_a_sse3(lv_32fc_t* re
|
|||
#ifdef LV_HAVE_SSE3
|
||||
static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_u_sse3(lv_32fc_t* result, const lv_32fc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 1023;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -135,8 +136,8 @@ static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_u_sse3(lv_32fc_t* re
|
|||
#ifdef LV_HAVE_SSE4_1
|
||||
static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_u_sse4_1(lv_32fc_t* result, const lv_32fc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 1023;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -164,8 +165,8 @@ static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_u_sse4_1(lv_32fc_t*
|
|||
#ifdef LV_HAVE_SSE4_1
|
||||
static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_a_sse4_1(lv_32fc_t* result, const lv_32fc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 1023;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -193,8 +194,8 @@ static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_a_sse4_1(lv_32fc_t*
|
|||
#ifdef LV_HAVE_AVX
|
||||
static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_a_avx(lv_32fc_t* result, const lv_32fc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 1023;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -222,8 +223,8 @@ static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_a_avx(lv_32fc_t* res
|
|||
#ifdef LV_HAVE_AVX
|
||||
static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_u_avx(lv_32fc_t* result, const lv_32fc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 1023;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -251,8 +252,8 @@ static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_u_avx(lv_32fc_t* res
|
|||
#ifdef LV_HAVE_AVX2
|
||||
static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_a_avx2(lv_32fc_t* result, const lv_32fc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 1023;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -280,8 +281,8 @@ static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_a_avx2(lv_32fc_t* re
|
|||
#ifdef LV_HAVE_AVX2
|
||||
static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_u_avx2(lv_32fc_t* result, const lv_32fc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 1023;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
@ -309,8 +310,8 @@ static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_u_avx2(lv_32fc_t* re
|
|||
#ifdef LV_HAVE_NEON
|
||||
static inline void volk_gnsssdr_32fc_resamplerxnpuppet_32fc_neon(lv_32fc_t* result, const lv_32fc_t* local_code, unsigned int num_points)
|
||||
{
|
||||
float code_phase_step_chips = -0.6;
|
||||
int code_length_chips = 1023;
|
||||
int code_length_chips = 2046;
|
||||
float code_phase_step_chips = ((float)(code_length_chips) + 0.1 )/( (float) num_points );
|
||||
int num_out_vectors = 3;
|
||||
float rem_code_phase_chips = -0.234;
|
||||
unsigned int n;
|
||||
|
|
|
|||
|
|
@ -89,10 +89,14 @@ std::vector<volk_gnsssdr_test_case_t> init_test_list(volk_gnsssdr_test_params_t
|
|||
(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_16i_resamplerxnpuppet_16i, volk_gnsssdr_16i_xn_resampler_16i_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_32f_resamplerxnpuppet_32f, volk_gnsssdr_32f_xn_resampler_32f_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))
|
||||
(VOLK_INIT_PUPP(volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic, volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn, test_params_int16))
|
||||
(VOLK_INIT_PUPP(volk_gnsssdr_32fc_x2_rotator_dotprodxnpuppet_32fc, volk_gnsssdr_32fc_x2_rotator_dot_prod_32fc_xn, test_params_int1))
|
||||
(VOLK_INIT_PUPP(volk_gnsssdr_32fc_32f_rotator_dotprodxnpuppet_32fc, volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn, test_params_int1))
|
||||
;
|
||||
|
||||
return test_cases;
|
||||
|
|
|
|||
|
|
@ -717,7 +717,7 @@ bool run_volk_gnsssdr_tests(volk_gnsssdr_func_desc_t desc,
|
|||
{
|
||||
if(both_sigs[j].is_signed)
|
||||
{
|
||||
fail = icompare((int8_t *) test_data[generic_offset][j], (int8_t *) test_data[i][j], vlen*(both_sigs[j].is_complex ? 2 : 1), tol_i);
|
||||
fail = icompare((int16_t *) test_data[generic_offset][j], (int16_t *) test_data[i][j], vlen*(both_sigs[j].is_complex ? 2 : 1), tol_i);
|
||||
}
|
||||
else
|
||||
{
|
||||
|
|
|
|||
Loading…
Reference in New Issue
Block a user