[hamradio-commits] [gnss-sdr] 138/236: Adding new resampler kernel and integrating it in the multicorrelator
Carles Fernandez
carles_fernandez-guest at moszumanska.debian.org
Tue Apr 26 16:02:45 UTC 2016
This is an automated email from the git hooks/post-receive script.
carles_fernandez-guest pushed a commit to branch next
in repository gnss-sdr.
commit 9eb175fb0e0ce037082b3879a70bf846fcf20da9
Author: Carles Fernandez <carles.fernandez at gmail.com>
Date: Wed Mar 30 21:33:43 2016 +0200
Adding new resampler kernel and integrating it in the multicorrelator
---
.../volk_gnsssdr_16ic_resamplerxnpuppet_16ic.h | 8 +-
.../volk_gnsssdr_32fc_resamplerxnpuppet_32fc.h | 248 ++++++++++++++++++
.../volk_gnsssdr_32fc_xn_resampler_32fc_xn.h | 290 +++++++++++++++++++++
.../tracking/libs/cpu_multicorrelator.cc | 35 +--
4 files changed, 556 insertions(+), 25 deletions(-)
diff --git a/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_resamplerxnpuppet_16ic.h b/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_resamplerxnpuppet_16ic.h
index 1563ad0..82a01be 100644
--- a/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_resamplerxnpuppet_16ic.h
+++ b/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_resamplerxnpuppet_16ic.h
@@ -45,7 +45,7 @@
static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_generic(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
{
float code_phase_step_chips = 0.1;
- int code_length_chips = 1023;
+ int code_length_chips = 2046;
int num_out_vectors = 3;
float* rem_code_phase_chips = (float*)volk_gnsssdr_malloc(sizeof(float) * num_out_vectors, volk_gnsssdr_get_alignment());
@@ -73,7 +73,7 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_generic(lv_16sc_t* r
static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_a_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
{
float code_phase_step_chips = 0.1;
- int code_length_chips = 1023;
+ int code_length_chips = 2046;
int num_out_vectors = 3;
float * rem_code_phase_chips = (float*)volk_gnsssdr_malloc(sizeof(float) * num_out_vectors, volk_gnsssdr_get_alignment());
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
@@ -100,7 +100,7 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_a_sse2(lv_16sc_t* re
static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_u_sse2(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
{
float code_phase_step_chips = 0.1;
- int code_length_chips = 1023;
+ int code_length_chips = 2046;
int num_out_vectors = 3;
float * rem_code_phase_chips = (float*)volk_gnsssdr_malloc(sizeof(float) * num_out_vectors, volk_gnsssdr_get_alignment());
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
@@ -127,7 +127,7 @@ static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_u_sse2(lv_16sc_t* re
static inline void volk_gnsssdr_16ic_resamplerxnpuppet_16ic_neon(lv_16sc_t* result, const lv_16sc_t* local_code, unsigned int num_points)
{
float code_phase_step_chips = 0.1;
- int code_length_chips = 1023;
+ int code_length_chips = 2046;
int num_out_vectors = 3;
float * rem_code_phase_chips = (float*)volk_gnsssdr_malloc(sizeof(float) * num_out_vectors, volk_gnsssdr_get_alignment());
lv_16sc_t** result_aux = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
diff --git a/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_32fc_resamplerxnpuppet_32fc.h b/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_32fc_resamplerxnpuppet_32fc.h
new file mode 100644
index 0000000..84ccd49
--- /dev/null
+++ b/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_32fc_resamplerxnpuppet_32fc.h
@@ -0,0 +1,248 @@
+/*!
+ * \file volk_gnsssdr_32fc_resamplerxnpuppet_32fc.h
+ * \brief VOLK_GNSSSDR puppet for the multiple 16-bit complex vector resampler kernel.
+ * \authors <ul>
+ * <li> Carles Fernandez Prades 2016 cfernandez at cttc dot cat
+ * </ul>
+ *
+ * VOLK_GNSSSDR puppet for integrating the multiple resampler into the test system
+ *
+ * -------------------------------------------------------------------------
+ *
+ * Copyright (C) 2010-2015 (see AUTHORS file for a list of contributors)
+ *
+ * GNSS-SDR is a software defined Global Navigation
+ * Satellite Systems receiver
+ *
+ * This file is part of GNSS-SDR.
+ *
+ * GNSS-SDR is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * GNSS-SDR is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with GNSS-SDR. If not, see <http://www.gnu.org/licenses/>.
+ *
+ * -------------------------------------------------------------------------
+ */
+
+#ifndef INCLUDED_volk_gnsssdr_32fc_resamplerxnpuppet_32fc_H
+#define INCLUDED_volk_gnsssdr_32fc_resamplerxnpuppet_32fc_H
+
+#include "volk_gnsssdr/volk_gnsssdr_32fc_xn_resampler_32fc_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_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 = 2046;
+ int num_out_vectors = 3;
+ float rem_code_phase_chips = -0.234;
+
+ float shifts_chips[3] = { -0.1, 0.0, 0.1 };
+
+ lv_32fc_t** result_aux = (lv_32fc_t**)volk_gnsssdr_malloc(sizeof(lv_32fc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ result_aux[n] = (lv_32fc_t*)volk_gnsssdr_malloc(sizeof(lv_32fc_t) * num_points, volk_gnsssdr_get_alignment());
+ }
+
+ volk_gnsssdr_32fc_xn_resampler_32fc_xn_generic(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
+
+ memcpy((lv_32fc_t*)result, (lv_32fc_t*)result_aux[0], sizeof(lv_32fc_t) * num_points);
+
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ volk_gnsssdr_free(result_aux[n]);
+ }
+ volk_gnsssdr_free(result_aux);
+}
+
+#endif /* LV_HAVE_GENERIC */
+
+
+#ifdef LV_HAVE_SSE3
+static inline void volk_gnsssdr_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 = 2046;
+ int num_out_vectors = 3;
+ float rem_code_phase_chips = -0.234;
+
+ float shifts_chips[3] = { -0.1, 0.0, 0.1 };
+
+ lv_32fc_t** result_aux = (lv_32fc_t**)volk_gnsssdr_malloc(sizeof(lv_32fc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ result_aux[n] = (lv_32fc_t*)volk_gnsssdr_malloc(sizeof(lv_32fc_t) * num_points, volk_gnsssdr_get_alignment());
+ }
+
+ volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_sse3(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
+
+ memcpy((lv_32fc_t*)result, (lv_32fc_t*)result_aux[0], sizeof(lv_32fc_t) * num_points);
+
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ volk_gnsssdr_free(result_aux[n]);
+ }
+ volk_gnsssdr_free(result_aux);
+}
+
+#endif
+
+#ifdef LV_HAVE_SSE3
+static inline void volk_gnsssdr_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 = 2046;
+ int num_out_vectors = 3;
+ float rem_code_phase_chips = -0.234;
+
+ float shifts_chips[3] = { -0.1, 0.0, 0.1 };
+
+ lv_32fc_t** result_aux = (lv_32fc_t**)volk_gnsssdr_malloc(sizeof(lv_32fc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ result_aux[n] = (lv_32fc_t*)volk_gnsssdr_malloc(sizeof(lv_32fc_t) * num_points, volk_gnsssdr_get_alignment());
+ }
+
+ volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_sse3(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
+
+ memcpy((lv_32fc_t*)result, (lv_32fc_t*)result_aux[0], sizeof(lv_32fc_t) * num_points);
+
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ volk_gnsssdr_free(result_aux[n]);
+ }
+ volk_gnsssdr_free(result_aux);
+}
+
+#endif
+
+
+#ifdef LV_HAVE_SSE4_1
+static inline void volk_gnsssdr_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 = 2046;
+ int num_out_vectors = 3;
+ float rem_code_phase_chips = -0.234;
+
+ float shifts_chips[3] = { -0.1, 0.0, 0.1 };
+
+ lv_32fc_t** result_aux = (lv_32fc_t**)volk_gnsssdr_malloc(sizeof(lv_32fc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ result_aux[n] = (lv_32fc_t*)volk_gnsssdr_malloc(sizeof(lv_32fc_t) * num_points, volk_gnsssdr_get_alignment());
+ }
+
+ volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_sse4_1(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
+
+ memcpy((lv_32fc_t*)result, (lv_32fc_t*)result_aux[0], sizeof(lv_32fc_t) * num_points);
+
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ volk_gnsssdr_free(result_aux[n]);
+ }
+ volk_gnsssdr_free(result_aux);
+}
+
+#endif
+
+#ifdef LV_HAVE_SSE4_1
+static inline void volk_gnsssdr_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 = 2046;
+ int num_out_vectors = 3;
+ float rem_code_phase_chips = -0.234;
+
+ float shifts_chips[3] = { -0.1, 0.0, 0.1 };
+
+ lv_32fc_t** result_aux = (lv_32fc_t**)volk_gnsssdr_malloc(sizeof(lv_32fc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ result_aux[n] = (lv_32fc_t*)volk_gnsssdr_malloc(sizeof(lv_32fc_t) * num_points, volk_gnsssdr_get_alignment());
+ }
+
+ volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_sse4_1(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
+
+ memcpy((lv_32fc_t*)result, (lv_32fc_t*)result_aux[0], sizeof(lv_32fc_t) * num_points);
+
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ volk_gnsssdr_free(result_aux[n]);
+ }
+ volk_gnsssdr_free(result_aux);
+}
+
+#endif
+
+#ifdef LV_HAVE_AVX
+static inline void volk_gnsssdr_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 = 2046;
+ int num_out_vectors = 3;
+ float rem_code_phase_chips = -0.234;
+
+ float shifts_chips[3] = { -0.1, 0.0, 0.1 };
+
+ lv_32fc_t** result_aux = (lv_32fc_t**)volk_gnsssdr_malloc(sizeof(lv_32fc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ result_aux[n] = (lv_32fc_t*)volk_gnsssdr_malloc(sizeof(lv_32fc_t) * num_points, volk_gnsssdr_get_alignment());
+ }
+
+ volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_avx(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
+
+ memcpy((lv_32fc_t*)result, (lv_32fc_t*)result_aux[0], sizeof(lv_32fc_t) * num_points);
+
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ volk_gnsssdr_free(result_aux[n]);
+ }
+ volk_gnsssdr_free(result_aux);
+}
+#endif
+
+
+#ifdef LV_HAVE_AVX
+static inline void volk_gnsssdr_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 = 2046;
+ int num_out_vectors = 3;
+ float rem_code_phase_chips = -0.234;
+
+ float shifts_chips[3] = { -0.1, 0.0, 0.1 };
+
+ lv_32fc_t** result_aux = (lv_32fc_t**)volk_gnsssdr_malloc(sizeof(lv_32fc_t*) * num_out_vectors, volk_gnsssdr_get_alignment());
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ result_aux[n] = (lv_32fc_t*)volk_gnsssdr_malloc(sizeof(lv_32fc_t) * num_points, volk_gnsssdr_get_alignment());
+ }
+
+ volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_avx(result_aux, local_code, rem_code_phase_chips, code_phase_step_chips, shifts_chips, code_length_chips, num_out_vectors, num_points);
+
+ memcpy((lv_32fc_t*)result, (lv_32fc_t*)result_aux[0], sizeof(lv_32fc_t) * num_points);
+
+ for(unsigned int n = 0; n < num_out_vectors; n++)
+ {
+ volk_gnsssdr_free(result_aux[n]);
+ }
+ volk_gnsssdr_free(result_aux);
+}
+#endif
+
+#endif // INCLUDED_volk_gnsssdr_32fc_resamplerpuppet_32fc_H
diff --git a/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_32fc_xn_resampler_32fc_xn.h b/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_32fc_xn_resampler_32fc_xn.h
new file mode 100644
index 0000000..6f9374e
--- /dev/null
+++ b/src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_32fc_xn_resampler_32fc_xn.h
@@ -0,0 +1,290 @@
+/*!
+ * \file volk_gnsssdr_16ic_xn_resampler_16ic_xn.h
+ * \brief VOLK_GNSSSDR kernel: Resamples N 16 bits integer short complex vectors using zero hold resample algorithm.
+ * \authors <ul>
+ * <li> Javier Arribas, 2015. jarribas(at)cttc.es
+ * </ul>
+ *
+ * VOLK_GNSSSDR kernel that esamples N 16 bits integer short complex vectors using zero hold resample algorithm.
+ * It is optimized to resample a sigle GNSS local code signal replica into N vectors fractional-resampled and fractional-delayed
+ * (i.e. it creates the Early, Prompt, and Late code replicas)
+ *
+ * -------------------------------------------------------------------------
+ *
+ * Copyright (C) 2010-2015 (see AUTHORS file for a list of contributors)
+ *
+ * GNSS-SDR is a software defined Global Navigation
+ * Satellite Systems receiver
+ *
+ * This file is part of GNSS-SDR.
+ *
+ * GNSS-SDR is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * GNSS-SDR is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with GNSS-SDR. If not, see <http://www.gnu.org/licenses/>.
+ *
+ * -------------------------------------------------------------------------
+ */
+
+/*!
+ * \page volk_gnsssdr_16ic_xn_resampler_16ic_xn
+ *
+ * \b Overview
+ *
+ * Resamples a complex vector (16-bit integer each component), providing \p num_out_vectors outputs.
+ *
+ * <b>Dispatcher Prototype</b>
+ * \code
+ * void volk_gnsssdr_16ic_xn_resampler_16ic_xn(lv_16sc_t** result, const lv_16sc_t* local_code, float* rem_code_phase_chips, float code_phase_step_chips, unsigned int code_length_chips, int num_out_vectors, unsigned int num_output_samples)
+ * \endcode
+ *
+ * \b Inputs
+ * \li local_code: One of the vectors to be multiplied.
+ * \li rem_code_phase_chips: Remnant code phase [chips].
+ * \li code_phase_step_chips: Phase increment per sample [chips/sample].
+ * \li code_length_chips: Code length in chips.
+ * \li num_out_vectors Number of output vectors.
+ * \li num_output_samples: The number of data values to be in the resampled vector.
+ *
+ * \b Outputs
+ * \li result: Pointer to a vector of pointers where the results will be stored.
+ *
+ */
+
+#ifndef INCLUDED_volk_gnsssdr_32fc_xn_resampler_32fc_xn_H
+#define INCLUDED_volk_gnsssdr_32fc_xn_resampler_32fc_xn_H
+
+#include <math.h>
+#include <volk_gnsssdr/volk_gnsssdr_common.h>
+#include <volk_gnsssdr/volk_gnsssdr_complex.h>
+
+//#pragma STDC FENV_ACCESS ON
+
+#ifdef LV_HAVE_GENERIC
+
+
+static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_generic(lv_32fc_t** result, const lv_32fc_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_output_samples)
+{
+ int local_code_chip_index;
+ for (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
+ {
+ for (int n = 0; n < num_output_samples; 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);
+ local_code_chip_index = local_code_chip_index % code_length_chips;
+ //Take into account that in multitap correlators, the shifts can be negative!
+ if (local_code_chip_index < 0) 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_32fc_xn_resampler_32fc_xn_a_sse3(lv_32fc_t** result, const lv_32fc_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_output_samples)
+{
+ lv_32fc_t** _result = result;
+ const unsigned int quarterPoints = num_output_samples / 4;
+
+ const __m128 ones = _mm_set1_ps(1.0f);
+ const __m128 fours = _mm_set1_ps(4.0f);
+ const __m128 rem_code_phase_chips_reg = _mm_set_ps1(rem_code_phase_chips);
+ const __m128 code_phase_step_chips_reg = _mm_set_ps1(code_phase_step_chips);
+
+ __attribute__((aligned(16))) int local_code_chip_index[4];
+ int local_code_chip_index_;
+
+ const __m128i zeros = _mm_setzero_si128();
+ const __m128 code_length_chips_reg_f = _mm_set_ps1((float)code_length_chips);
+ const __m128i code_length_chips_reg_i = _mm_set1_epi32((int)code_length_chips);
+ __m128i local_code_chip_index_reg, aux_i, negatives, i;
+ __m128 aux, aux2, shifts_chips_reg, fi, igx, j, c, cTrunc, base;
+
+ for (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
+ {
+ shifts_chips_reg = _mm_set_ps1((float)shifts_chips[current_correlator_tap]);
+ aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
+ __m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
+ for(unsigned int n = 0; n < quarterPoints; n++)
+ {
+ aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
+ aux = _mm_add_ps(aux, aux2);
+ // floor
+ i = _mm_cvttps_epi32(aux);
+ fi = _mm_cvtepi32_ps(i);
+ igx = _mm_cmpgt_ps(fi, aux);
+ j = _mm_and_ps(igx, ones);
+ aux = _mm_sub_ps(fi, j);
+ // fmod
+ c = _mm_div_ps(aux, code_length_chips_reg_f);
+ i = _mm_cvttps_epi32(c);
+ cTrunc = _mm_cvtepi32_ps(i);
+ base = _mm_mul_ps(cTrunc, code_length_chips_reg_f);
+ local_code_chip_index_reg = _mm_cvtps_epi32(_mm_sub_ps(aux, base));
+
+ negatives = _mm_cmplt_epi32(local_code_chip_index_reg, zeros);
+ aux_i = _mm_and_si128(code_length_chips_reg_i, negatives);
+ local_code_chip_index_reg = _mm_add_epi32(local_code_chip_index_reg, aux_i);
+ _mm_store_si128((__m128i*)local_code_chip_index, local_code_chip_index_reg);
+ for(unsigned int k = 0; k < 4; ++k)
+ {
+ _result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
+ }
+ indexn = _mm_add_ps(indexn, fours);
+ }
+ for(unsigned int n = quarterPoints * 4; n < num_output_samples; 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);
+ local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
+ //Take into account that in multitap correlators, the shifts can be negative!
+ if (local_code_chip_index_ < 0) 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_32fc_xn_resampler_32fc_xn_a_sse4_1(lv_32fc_t** result, const lv_32fc_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_output_samples)
+{
+ lv_32fc_t** _result = result;
+ const unsigned int quarterPoints = num_output_samples / 4;
+
+ const __m128 fours = _mm_set1_ps(4.0f);
+ const __m128 rem_code_phase_chips_reg = _mm_set_ps1(rem_code_phase_chips);
+ const __m128 code_phase_step_chips_reg = _mm_set_ps1(code_phase_step_chips);
+
+ __attribute__((aligned(16))) int local_code_chip_index[4];
+ int local_code_chip_index_;
+
+ const __m128i zeros = _mm_setzero_si128();
+ const __m128 code_length_chips_reg_f = _mm_set_ps1((float)code_length_chips);
+ const __m128i code_length_chips_reg_i = _mm_set1_epi32((int)code_length_chips);
+ __m128i local_code_chip_index_reg, aux_i, negatives, i;
+ __m128 aux, aux2, shifts_chips_reg, c, cTrunc, base;
+
+ for (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
+ {
+ shifts_chips_reg = _mm_set_ps1((float)shifts_chips[current_correlator_tap]);
+ aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
+ __m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
+ for(unsigned int n = 0; n < quarterPoints; n++)
+ {
+ aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
+ aux = _mm_add_ps(aux, aux2);
+ // floor
+ aux = _mm_floor_ps(aux);
+
+ // fmod
+ c = _mm_div_ps(aux, code_length_chips_reg_f);
+ i = _mm_cvttps_epi32(c);
+ cTrunc = _mm_cvtepi32_ps(i);
+ base = _mm_mul_ps(cTrunc, code_length_chips_reg_f);
+ local_code_chip_index_reg = _mm_cvtps_epi32(_mm_sub_ps(aux, base));
+
+ negatives = _mm_cmplt_epi32(local_code_chip_index_reg, zeros);
+ aux_i = _mm_and_si128(code_length_chips_reg_i, negatives);
+ local_code_chip_index_reg = _mm_add_epi32(local_code_chip_index_reg, aux_i);
+ _mm_store_si128((__m128i*)local_code_chip_index, local_code_chip_index_reg);
+ for(unsigned int k = 0; k < 4; ++k)
+ {
+ _result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
+ }
+ indexn = _mm_add_ps(indexn, fours);
+ }
+ for(unsigned int n = quarterPoints * 4; n < num_output_samples; 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);
+ local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
+ //Take into account that in multitap correlators, the shifts can be negative!
+ if (local_code_chip_index_ < 0) 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_32fc_xn_resampler_32fc_xn_a_avx(lv_32fc_t** result, const lv_32fc_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_output_samples)
+{
+ lv_32fc_t** _result = result;
+ const unsigned int avx_iters = num_output_samples / 8;
+
+ const __m256 eights = _mm256_set1_ps(8.0f);
+ const __m256 rem_code_phase_chips_reg = _mm256_set1_ps(rem_code_phase_chips);
+ const __m256 code_phase_step_chips_reg = _mm256_set1_ps(code_phase_step_chips);
+
+ __attribute__((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);
+
+ __m256i local_code_chip_index_reg, i;
+ __m256 aux, aux2, shifts_chips_reg, c, cTrunc, base, negatives;
+
+ for (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
+ {
+ shifts_chips_reg = _mm256_set1_ps((float)shifts_chips[current_correlator_tap]);
+ aux2 = _mm256_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
+ __m256 indexn = _mm256_set_ps(7.0f, 6.0f, 5.0f, 4.0f, 3.0f, 2.0f, 1.0f, 0.0f);
+ for(unsigned int n = 0; n < avx_iters; n++)
+ {
+ 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);
+ aux = _mm256_sub_ps(aux, base);
+
+ negatives = _mm256_cmp_ps(aux, zeros, 0x01);
+ aux2 = _mm256_and_ps(code_length_chips_reg_f, negatives);
+ local_code_chip_index_reg = _mm256_cvtps_epi32(_mm256_add_ps(aux, aux2));
+ _mm256_store_si256((__m256i*)local_code_chip_index, local_code_chip_index_reg);
+ for(unsigned int k = 0; k < 8; ++k)
+ {
+ _result[current_correlator_tap][n * 8 + k] = local_code[local_code_chip_index[k]];
+ }
+ indexn = _mm256_add_ps(indexn, eights);
+ }
+ _mm256_zeroupper();
+ for(unsigned int n = avx_iters * 8; n < num_output_samples; 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);
+ local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
+ //Take into account that in multitap correlators, the shifts can be negative!
+ if (local_code_chip_index_ < 0) local_code_chip_index_ += code_length_chips;
+ _result[current_correlator_tap][n] = local_code[local_code_chip_index_];
+ }
+
+ }
+}
+#endif
+
+#endif /*INCLUDED_volk_gnsssdr_16ic_xn_resampler_16ic_xn_H*/
+
diff --git a/src/algorithms/tracking/libs/cpu_multicorrelator.cc b/src/algorithms/tracking/libs/cpu_multicorrelator.cc
index 751ad22..81cdfc7 100644
--- a/src/algorithms/tracking/libs/cpu_multicorrelator.cc
+++ b/src/algorithms/tracking/libs/cpu_multicorrelator.cc
@@ -35,7 +35,6 @@
#include "cpu_multicorrelator.h"
#include <cmath>
#include <iostream>
-#include <volk/volk.h>
#include <volk_gnsssdr/volk_gnsssdr.h>
@@ -68,10 +67,10 @@ bool cpu_multicorrelator::init(
// ALLOCATE MEMORY FOR INTERNAL vectors
size_t size = max_signal_length_samples * sizeof(std::complex<float>);
- d_local_codes_resampled = static_cast<std::complex<float>**>(volk_malloc(n_correlators * sizeof(std::complex<float>), volk_get_alignment()));
+ d_local_codes_resampled = static_cast<std::complex<float>**>(volk_gnsssdr_malloc(n_correlators * sizeof(std::complex<float>), volk_gnsssdr_get_alignment()));
for (int n = 0; n < n_correlators; n++)
{
- d_local_codes_resampled[n] = static_cast<std::complex<float>*>(volk_malloc(size, volk_get_alignment()));
+ d_local_codes_resampled[n] = static_cast<std::complex<float>*>(volk_gnsssdr_malloc(size, volk_gnsssdr_get_alignment()));
}
d_n_correlators = n_correlators;
return true;
@@ -100,23 +99,17 @@ bool cpu_multicorrelator::set_input_output_vectors(std::complex<float>* corr_out
return true;
}
-void cpu_multicorrelator::update_local_code(int correlator_length_samples,float rem_code_phase_chips, float code_phase_step_chips)
-{
- int local_code_chip_index;
- for (int current_correlator_tap = 0; current_correlator_tap < d_n_correlators; current_correlator_tap++)
-
- {
- for (int n = 0; n < correlator_length_samples; n++)
- {
- // resample code for current tap
- local_code_chip_index = floor(code_phase_step_chips*static_cast<float>(n) + d_shifts_chips[current_correlator_tap]- rem_code_phase_chips);
- local_code_chip_index = local_code_chip_index % d_code_length_chips;
- //Take into account that in multitap correlators, the shifts can be negative!
- if (local_code_chip_index < 0) local_code_chip_index += d_code_length_chips;
- d_local_codes_resampled[current_correlator_tap][n] = d_local_code_in[local_code_chip_index];
- }
- }
+void cpu_multicorrelator::update_local_code(int correlator_length_samples, float rem_code_phase_chips, float code_phase_step_chips)
+{
+ volk_gnsssdr_32fc_xn_resampler_32fc_xn(d_local_codes_resampled,
+ d_local_code_in,
+ rem_code_phase_chips,
+ code_phase_step_chips,
+ d_shifts_chips,
+ correlator_length_samples,
+ d_n_correlators,
+ d_code_length_chips);
}
@@ -142,9 +135,9 @@ bool cpu_multicorrelator::free()
// Free memory
for (int n = 0; n < d_n_correlators; n++)
{
- volk_free(d_local_codes_resampled[n]);
+ volk_gnsssdr_free(d_local_codes_resampled[n]);
}
- volk_free(d_local_codes_resampled);
+ volk_gnsssdr_free(d_local_codes_resampled);
return true;
}
--
Alioth's /usr/local/bin/git-commit-notice on /srv/git.debian.org/git/pkg-hamradio/gnss-sdr.git
More information about the pkg-hamradio-commits
mailing list