mirrors/gnss-sdr
1
0
Fork 0
mirror of https://github.com/gnss-sdr/gnss-sdr synced 2024-06-24 22:13:15 +00:00
Code Issues Releases Wiki Activity

Integrating a new volk kernel

Browse Source
This commit is contained in:
Carles Fernandez 2016-01-13 11:42:01 +01:00
parent f659005e63
commit ece7bc2c65
3 changed files with 29 additions and 35 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

35
src/algorithms/tracking/libs/volk_gnsssdr_16ic_x2_multiply_16ic.h → src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_multiply_16ic.h
View File

@ -51,11 +51,11 @@
*/
static inline void volk_gnsssdr_16ic_x2_multiply_16ic_generic(lv_16sc_t* result, const lv_16sc_t* in_a, const lv_16sc_t* in_b, unsigned int num_points)
{
for (unsigned int n=0;n<num_points;n++)
{
//r*a.r - i*a.i, i*a.r + r*a.i
result[n]=in_a[n]*in_b[n];
}
for (unsigned int n = 0; n < num_points; n++)
{
//r*a.r - i*a.i, i*a.r + r*a.i
result[n] = in_a[n] * in_b[n];
}
}
#endif /*LV_HAVE_GENERIC*/
@ -74,27 +74,27 @@ static inline void volk_gnsssdr_16ic_x2_multiply_16ic_a_sse2(lv_16sc_t* out, con
const lv_16sc_t* _in_a = in_a;
const lv_16sc_t* _in_b = in_b;
lv_16sc_t* _out = out;
for(unsigned int number = 0;number < sse_iters; number++)
for(unsigned int number = 0; number < sse_iters; number++)
{
//std::complex<T> memory structure: real part -> reinterpret_cast<cv T*>(a)[2*i]
//imaginery part -> reinterpret_cast<cv T*>(a)[2*i + 1]
// a[127:0]=[a3.i,a3.r,a2.i,a2.r,a1.i,a1.r,a0.i,a0.r]
//std::complex<T> memory structure: real part -> reinterpret_cast<cv T*>(a)[2*i]
//imaginery part -> reinterpret_cast<cv T*>(a)[2*i + 1]
// a[127:0]=[a3.i,a3.r,a2.i,a2.r,a1.i,a1.r,a0.i,a0.r]
a = _mm_loadu_si128((__m128i*)_in_a); //load (2 byte imag, 2 byte real) x 4 into 128 bits reg
b = _mm_loadu_si128((__m128i*)_in_b);
c=_mm_mullo_epi16 (a, b); // a3.i*b3.i, a3.r*b3.r, ....
c = _mm_mullo_epi16 (a, b); // a3.i*b3.i, a3.r*b3.r, ....
c_sr = _mm_srli_si128 (c, 2); // Shift a right by imm8 bytes while shifting in zeros, and store the results in dst.
real = _mm_subs_epi16 (c,c_sr);
real = _mm_subs_epi16 (c, c_sr);
real = _mm_and_si128 (real, mask_real); // a3.r*b3.r-a3.i*b3.i , 0, a3.r*b3.r- a3.i*b3.i
b_sl = _mm_slli_si128(b,2); // b3.r, b2.i ....
a_sl = _mm_slli_si128(a,2); // a3.r, a2.i ....
b_sl = _mm_slli_si128(b, 2); // b3.r, b2.i ....
a_sl = _mm_slli_si128(a, 2); // a3.r, a2.i ....
imag1 = _mm_mullo_epi16(a,b_sl); // a3.i*b3.r, ....
imag2 = _mm_mullo_epi16(b,a_sl); // b3.i*a3.r, ....
imag1 = _mm_mullo_epi16(a, b_sl); // a3.i*b3.r, ....
imag2 = _mm_mullo_epi16(b, a_sl); // b3.i*a3.r, ....
imag = _mm_adds_epi16(imag1,imag2);
imag = _mm_adds_epi16(imag1, imag2);
imag = _mm_and_si128 (imag, mask_imag); // a3.i*b3.r+b3.i*a3.r, 0, ...
result = _mm_or_si128 (real, imag);
@ -104,10 +104,9 @@ static inline void volk_gnsssdr_16ic_x2_multiply_16ic_a_sse2(lv_16sc_t* out, con
_in_a += 4;
_in_b += 4;
_out += 4;
}
for (unsigned int i = 0; i<(num_points % 4); ++i)
for (unsigned int i = 0; i < (num_points % 4); ++i)
{
*_out++ = (*_in_a++) * (*_in_b++);
}

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

@ -80,6 +80,7 @@ std::vector<volk_gnsssdr_test_case_t> init_test_list(volk_gnsssdr_test_params_t
(VOLK_INIT_TEST(volk_gnsssdr_32fc_convert_8ic, test_params))
(VOLK_INIT_TEST(volk_gnsssdr_32fc_convert_16ic, test_params))
(VOLK_INIT_TEST(volk_gnsssdr_16ic_x2_dot_prod_16ic, test_params))
(VOLK_INIT_TEST(volk_gnsssdr_16ic_x2_multiply_16ic, test_params))
//(VOLK_INIT_TEST(volk_gnsssdr_16ic_resampler_16ic, test_params))
;

28
src/algorithms/tracking/libs/cpu_multicorrelator_16sc.cc
View File

@ -41,19 +41,14 @@
#define LV_HAVE_GENERIC
#define LV_HAVE_SSE2
//#include "volk_gnsssdr_16ic_x2_dot_prod_16ic.h"
#include "volk_gnsssdr_16ic_x2_multiply_16ic.h"
//#include "volk_gnsssdr_16ic_resampler_16ic.h"
#include "volk_gnsssdr_16ic_xn_resampler_16ic_xn.h"
#include "volk_gnsssdr_16ic_xn_dot_prod_16ic_xn.h"
bool cpu_multicorrelator_16sc::init(
int max_signal_length_samples,
int n_correlators
)
int max_signal_length_samples,
int n_correlators)
{
// ALLOCATE MEMORY FOR INTERNAL vectors
// ALLOCATE MEMORY FOR INTERNAL vectors
size_t size = max_signal_length_samples * sizeof(lv_16sc_t);
// NCO signal
@ -76,8 +71,7 @@ bool cpu_multicorrelator_16sc::init(
bool cpu_multicorrelator_16sc::set_local_code_and_taps(
int code_length_chips,
const lv_16sc_t* local_code_in,
float *shifts_chips
)
float *shifts_chips)
{
d_local_code_in = local_code_in;
d_shifts_chips = shifts_chips;
@ -98,12 +92,11 @@ bool cpu_multicorrelator_16sc::set_input_output_vectors(lv_16sc_t* corr_out, con
void cpu_multicorrelator_16sc::update_local_code(int correlator_length_samples,float rem_code_phase_chips, float code_phase_step_chips)
{
float *tmp_code_phases_chips;
tmp_code_phases_chips=static_cast<float*>(volk_malloc(d_n_correlators*sizeof(float), volk_get_alignment()));
for (int n=0;n<d_n_correlators;n++)
tmp_code_phases_chips = static_cast<float*>(volk_malloc(d_n_correlators*sizeof(float), volk_get_alignment()));
for (int n = 0; n < d_n_correlators; n++)
{
tmp_code_phases_chips[n]=d_shifts_chips[n]-rem_code_phase_chips;
tmp_code_phases_chips[n] = d_shifts_chips[n] - rem_code_phase_chips;
}
volk_gnsssdr_16ic_xn_resampler_16ic_xn_sse2(d_local_codes_resampled,
@ -155,11 +148,12 @@ bool cpu_multicorrelator_16sc::Carrier_wipeoff_multicorrelator_resampler(
update_local_carrier(signal_length_samples, rem_carrier_phase_in_rad, phase_step_rad);
//std::cout<<"d_nco_in 16sc="<<d_nco_in[23]<<std::endl;
volk_gnsssdr_16ic_x2_multiply_16ic_a_sse2(d_sig_doppler_wiped,d_sig_in,d_nco_in,signal_length_samples);
//volk_gnsssdr_16ic_x2_multiply_16ic_a_sse2(d_sig_doppler_wiped,d_sig_in,d_nco_in,signal_length_samples);
volk_gnsssdr_16ic_x2_multiply_16ic(d_sig_doppler_wiped, d_sig_in, d_nco_in, signal_length_samples);
//std::cout<<"d_sig_doppler_wiped 16sc="<<d_sig_doppler_wiped[23]<<std::endl;
update_local_code(signal_length_samples,rem_code_phase_chips, code_phase_step_chips);
update_local_code(signal_length_samples, rem_code_phase_chips, code_phase_step_chips);
volk_gnsssdr_16ic_xn_dot_prod_16ic_xn_a_sse2(d_corr_out, d_sig_doppler_wiped, (const lv_16sc_t**)d_local_codes_resampled,signal_length_samples,d_n_correlators);
volk_gnsssdr_16ic_xn_dot_prod_16ic_xn_a_sse2(d_corr_out, d_sig_doppler_wiped, (const lv_16sc_t**)d_local_codes_resampled, signal_length_samples, d_n_correlators);
//for (int current_correlator_tap = 0; current_correlator_tap < d_n_correlators; current_correlator_tap++)
// {