mirrors/gnss-sdr
1
0
Fork 0
mirror of https://github.com/gnss-sdr/gnss-sdr synced 2025-02-21 05:20:37 +00:00
Code Issues Releases Wiki Activity

Improve generic kernel

Browse Source
This commit is contained in:
Antonio Ramos 2018-09-06 17:46:55 +02:00
parent 4792bc32c6
commit 47e5ef7f39
6 changed files with 706 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

512
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_32fc_32f_high_dynamic_rotator_dot_prod_32fc_xn.h Normal file
View File

@ -0,0 +1,512 @@
/*!
* \file volk_gnsssdr_32fc_32f_high_dynamic_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) 2010-2018 (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 <https://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_high_dynamic_rotator_dot_prod_32fc_xn_H
#define INCLUDED_volk_gnsssdr_32fc_32f_high_dynamic_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_high_dynamic_rotator_dot_prod_32fc_xn_generic(lv_32fc_t* result, const lv_32fc_t* in_common, const lv_32fc_t phase_inc, const lv_32fc_t phase_inc_rate, lv_32fc_t* phase, const float** in_a, int num_a_vectors, unsigned int num_points)
{
lv_32fc_t tmp32_1, tmp32_2;
#ifdef __cplusplus
float d_arg = std::arg(phase_inc_rate) / 2.0;
// float d_arg = 0.5; //BORRAR
lv_32fc_t half_phase_inc_rate = std::exp(0.0, d_arg);
#else
float d_arg = carg(phase_inc_rate) / 2.0;
// float d_arg = 0.5; //BORRAR
lv_32fc_t half_phase_inc_rate = cexp(lv_cmake(0.0, d_arg));
#endif
// lv_32fc_t half_phase_inc_rate = lv_cmake(0.0, 1.0); //BORRAR
lv_32fc_t delta_phase_rate = lv_cmake(1.0, 0.0);
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));
delta_phase_rate /= std::abs(delta_phase_rate);
#else
(*phase) /= hypotf(lv_creal(*phase), lv_cimag(*phase));
delta_phase_rate /= hypotf(lv_creal(delta_phase_rate), lv_cimag(delta_phase_rate));
#endif
//printf("Phase after regeneration %i: %f,%f Modulus: %f\n", n,lv_creal(*phase),lv_cimag(*phase), cabsf(*phase));
}
(*phase) *= (phase_inc * half_phase_inc_rate * delta_phase_rate);
delta_phase_rate *= phase_inc_rate;
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;
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;
__VOLK_ATTR_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
#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_a_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;
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;
__VOLK_ATTR_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_load_ps(aPtr);
a1Val = _mm256_load_ps(aPtr + 8);
a2Val = _mm256_load_ps(aPtr + 16);
a3Val = _mm256_load_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_high_dynamic_rotator_dot_prod_32fc_xn_H */

163
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_32fc_32f_high_dynamic_rotator_dotprodxnpuppet_32fc.h Normal file
View File

@ -0,0 +1,163 @@
/*!
* \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) 2010-2018 (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 <https://www.gnu.org/licenses/>.
*
* -------------------------------------------------------------------------
*/
#ifndef INCLUDED_volk_gnsssdr_32fc_32f_high_dynamic_rotator_dotprodxnpuppet_32fc_H
#define INCLUDED_volk_gnsssdr_32fc_32f_high_dynamic_rotator_dotprodxnpuppet_32fc_H
#include "volk_gnsssdr/volk_gnsssdr_32fc_32f_high_dynamic_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_high_dynamic_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));
lv_32fc_t phase_inc_rate[1];
phase_inc_rate[0] = lv_cmake(cos(phase_step_rad * 0.001), sin(phase_step_rad * 0.001));
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_high_dynamic_rotator_dot_prod_32fc_xn_generic(result, local_code, phase_inc[0], phase_inc_rate[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));
// 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));
// 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
//
//
//#ifdef LV_HAVE_AVX
//static inline void volk_gnsssdr_32fc_32f_rotator_dotprodxnpuppet_32fc_a_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));
// 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_a_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_high_dynamic_rotator_dotprodxnpuppet_32fc_H

1
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/lib/kernel_tests.h
View File

@ -99,6 +99,7 @@ std::vector<volk_gnsssdr_test_case_t> init_test_list(volk_gnsssdr_test_params_t
QA(VOLK_INIT_PUPP(volk_gnsssdr_16ic_16i_rotator_dotprodxnpuppet_16ic, volk_gnsssdr_16ic_16i_rotator_dot_prod_16ic_xn, test_params_int16))
QA(VOLK_INIT_PUPP(volk_gnsssdr_32fc_x2_rotator_dotprodxnpuppet_32fc, volk_gnsssdr_32fc_x2_rotator_dot_prod_32fc_xn, test_params_int1))
QA(VOLK_INIT_PUPP(volk_gnsssdr_32fc_32f_rotator_dotprodxnpuppet_32fc, volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn, test_params_int1));
QA(VOLK_INIT_PUPP(volk_gnsssdr_32fc_32f_high_dynamic_rotator_dotprodxnpuppet_32fc, volk_gnsssdr_32fc_32f_high_dynamic_rotator_dot_prod_32fc_xn, test_params_int1));
return test_cases;
}

4
src/algorithms/tracking/gnuradio_blocks/dll_pll_veml_tracking.cc
View File

@ -710,7 +710,7 @@ void dll_pll_veml_tracking::do_correlation_step(const gr_complex *input_samples)
multicorrelator_cpu.set_input_output_vectors(d_correlator_outs, input_samples);
multicorrelator_cpu.Carrier_wipeoff_multicorrelator_resampler(
d_rem_carr_phase_rad,
d_carrier_phase_step_rad,
d_carrier_phase_step_rad, d_carrier_phase_rate_step_rad,
static_cast<float>(d_rem_code_phase_chips) * static_cast<float>(d_code_samples_per_chip),
static_cast<float>(d_code_phase_step_chips) * static_cast<float>(d_code_samples_per_chip),
static_cast<float>(d_code_phase_rate_step_chips) * static_cast<float>(d_code_samples_per_chip),
@ -722,7 +722,7 @@ void dll_pll_veml_tracking::do_correlation_step(const gr_complex *input_samples)
correlator_data_cpu.set_input_output_vectors(d_Prompt_Data, input_samples);
correlator_data_cpu.Carrier_wipeoff_multicorrelator_resampler(
d_rem_carr_phase_rad,
d_carrier_phase_step_rad,
d_carrier_phase_step_rad, d_carrier_phase_rate_step_rad,
static_cast<float>(d_rem_code_phase_chips) * static_cast<float>(d_code_samples_per_chip),
static_cast<float>(d_code_phase_step_chips) * static_cast<float>(d_code_samples_per_chip),
static_cast<float>(d_code_phase_rate_step_chips) * static_cast<float>(d_code_samples_per_chip),

27
src/algorithms/tracking/libs/cpu_multicorrelator_real_codes.cc
View File

@ -125,7 +125,32 @@ void cpu_multicorrelator_real_codes::update_local_code(int correlator_length_sam
}
}
// Overload Carrier_wipeoff_multicorrelator_resampler to ensure back compatibility
bool cpu_multicorrelator_real_codes::Carrier_wipeoff_multicorrelator_resampler(
float rem_carrier_phase_in_rad,
float phase_step_rad,
float phase_rate_step_rad,
float rem_code_phase_chips,
float code_phase_step_chips,
float code_phase_rate_step_chips,
int signal_length_samples)
{
update_local_code(signal_length_samples, rem_code_phase_chips, code_phase_step_chips, code_phase_rate_step_chips);
// Regenerate phase at each call in order to avoid numerical issues
lv_32fc_t phase_offset_as_complex[1];
phase_offset_as_complex[0] = lv_cmake(std::cos(rem_carrier_phase_in_rad), -std::sin(rem_carrier_phase_in_rad));
// call VOLK_GNSSSDR kernel
if (d_use_high_dynamics_resampler)
{
volk_gnsssdr_32fc_32f_high_dynamic_rotator_dot_prod_32fc_xn(d_corr_out, d_sig_in, std::exp(lv_32fc_t(0.0, -phase_step_rad)), std::exp(lv_32fc_t(0.0, -phase_rate_step_rad)), phase_offset_as_complex, const_cast<const float**>(d_local_codes_resampled), d_n_correlators, signal_length_samples);
}
else
{
volk_gnsssdr_32fc_32f_rotator_dot_prod_32fc_xn(d_corr_out, d_sig_in, std::exp(lv_32fc_t(0.0, -phase_step_rad)), phase_offset_as_complex, const_cast<const float**>(d_local_codes_resampled), d_n_correlators, signal_length_samples);
}
return true;
}
// Overload Carrier_wipeoff_multicorrelator_resampler to ensure back compatibility
bool cpu_multicorrelator_real_codes::Carrier_wipeoff_multicorrelator_resampler(
float rem_carrier_phase_in_rad,
float phase_step_rad,

2
src/algorithms/tracking/libs/cpu_multicorrelator_real_codes.h
View File

@ -52,6 +52,8 @@ public:
bool set_local_code_and_taps(int code_length_chips, const float *local_code_in, float *shifts_chips);
bool set_input_output_vectors(std::complex<float> *corr_out, const std::complex<float> *sig_in);
void update_local_code(int correlator_length_samples, float rem_code_phase_chips, float code_phase_step_chips, float code_phase_rate_step_chips = 0.0);
// Overload Carrier_wipeoff_multicorrelator_resampler to ensure back compatibility
bool Carrier_wipeoff_multicorrelator_resampler(float rem_carrier_phase_in_rad, float phase_step_rad, float phase_rate_step_rad, float rem_code_phase_chips, float code_phase_step_chips, float code_phase_rate_step_chips, int signal_length_samples);
bool Carrier_wipeoff_multicorrelator_resampler(float rem_carrier_phase_in_rad, float phase_step_rad, float rem_code_phase_chips, float code_phase_step_chips, float code_phase_rate_step_chips, int signal_length_samples);
bool free();