mirrors/gnss-sdr
1
0
Fork 0
mirror of https://github.com/gnss-sdr/gnss-sdr synced 2025-03-13 23:18:15 +00:00
Code Issues Releases Wiki Activity

fixes

Browse Source
This commit is contained in:
Carles Fernandez 2016-02-19 08:25:41 +01:00
parent 505b13a21a
commit 39ce43f7e3
4 changed files with 48 additions and 38 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_dot_prod_16ic_xn.h
View File

@ -302,7 +302,7 @@ static inline void volk_gnsssdr_16ic_x2_dot_prod_16ic_xn_neon(lv_16sc_t* result,
//todo dyn mem reg
int16x4x2_t* accumulator;
accumulator = (int16x4x2_t*)calloc(num_a_vectors, sizeof(int16x4x2_t));
accumulator = (int16x4x2_t*)malloc(num_a_vectors * sizeof(int16x4x2_t));
int16x4x2_t tmp_real, tmp_imag;
@ -333,11 +333,11 @@ static inline void volk_gnsssdr_16ic_x2_dot_prod_16ic_xn_neon(lv_16sc_t* result,
// a0i*b0r|a1i*b1r|a2i*b2r|a3i*b3r
tmp_imag.val[1] = vmul_s16(a_val.val[1], b_val.val[0]);
c_val.val[0] = vsub_s16(tmp_real.val[0], tmp_real.val[1]);
c_val.val[1] = vadd_s16(tmp_imag.val[0], tmp_imag.val[1]);
c_val.val[0] = vqsub_s16(tmp_real.val[0], tmp_real.val[1]);
c_val.val[1] = vqadd_s16(tmp_imag.val[0], tmp_imag.val[1]);
accumulator[n_vec].val[0] = vadd_s16(accumulator[n_vec].val[0], c_val.val[0]);
accumulator[n_vec].val[1] = vadd_s16(accumulator[n_vec].val[1], c_val.val[1]);
accumulator[n_vec].val[0] = vqadd_s16(accumulator[n_vec].val[0], c_val.val[0]);
accumulator[n_vec].val[1] = vqadd_s16(accumulator[n_vec].val[1], c_val.val[1]);
}
_in_common += 4;
}

51
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn.h
View File

@ -43,6 +43,7 @@
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
#include <volk_gnsssdr/saturation_arithmetic.h>
#include <math.h>
#include <stdio.h>
#ifdef LV_HAVE_GENERIC
/*!
@ -65,13 +66,14 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_generic(lv_16sc
}
for (unsigned int n = 0; n < num_points; n++)
{
tmp16 = *in_common++;
tmp16 = *in_common++; if(n<10 || n >= 8108) printf("generic phase %i: %f,%f\n", n,lv_creal(*phase),lv_cimag(*phase));
tmp32 = lv_cmake((float)lv_creal(tmp16), (float)lv_cimag(tmp16)) * (*phase);
tmp16 = lv_cmake((int16_t)rintf(lv_creal(tmp32)), (int16_t)rintf(lv_cimag(tmp32)));
(*phase) *= phase_inc;
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
{
lv_16sc_t tmp = tmp16 * in_a[n_vec][n];
//lv_16sc_t tmp = lv_cmake(sat_adds16i(sat_muls16i(lv_creal(tmp16), lv_creal(in_a[n_vec][n])), - sat_muls16i(lv_cimag(tmp16), lv_cimag(in_a[n_vec][n]))) , sat_adds16i(sat_muls16i(lv_creal(tmp16), lv_cimag(in_a[n_vec][n])), sat_muls16i(lv_cimag(tmp16), lv_creal(in_a[n_vec][n]))));
result[n_vec] = lv_cmake(sat_adds16i(lv_creal(result[n_vec]), lv_creal(tmp)), sat_adds16i(lv_cimag(result[n_vec]), lv_cimag(tmp)));
}
}
@ -128,6 +130,8 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_a_sse3(lv_16sc_
__attribute__((aligned(16))) lv_32fc_t two_phase_acc[2];
two_phase_acc[0] = (*phase);
two_phase_acc[1] = (*phase) * phase_inc;
printf("a_sse phase %i: %f,%f\n", 0,lv_creal(two_phase_acc[0]),lv_cimag(two_phase_acc[0]));
printf("a_sse phase %i: %f,%f\n", 1,lv_creal(two_phase_acc[1]),lv_cimag(two_phase_acc[1]));
two_phase_acc_reg = _mm_load_ps((float*)two_phase_acc);
__m128 yl, yh, tmp1, tmp2, tmp3;
lv_16sc_t tmp16;
@ -136,7 +140,7 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_a_sse3(lv_16sc_
for(unsigned int number = 0; number < sse_iters; number++)
{
// Phase rotation on operand in_common starts here:
//printf("generic phase %i: %f,%f\n", n*4,lv_creal(*phase),lv_cimag(*phase));
pa = _mm_set_ps((float)(lv_cimag(_in_common[1])), (float)(lv_creal(_in_common[1])), (float)(lv_cimag(_in_common[0])), (float)(lv_creal(_in_common[0]))); // //load (2 byte imag, 2 byte real) x 2 into 128 bits reg
//complex 32fc multiplication b=a*two_phase_acc_reg
yl = _mm_moveldup_ps(two_phase_acc_reg); // Load yl with cr,cr,dr,dr
@ -223,17 +227,20 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_a_sse3(lv_16sc_
free(imagcacc);
_mm_store_ps((float*)two_phase_acc, two_phase_acc_reg);
(*phase) = lv_cmake(two_phase_acc[0], two_phase_acc[0]);
//(*phase) = lv_cmake((float*)two_phase_acc[0], (float*)two_phase_acc[1]);
(*phase) = two_phase_acc[0];
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
for(unsigned int n = sse_iters * 4; n < num_points; n++)
{
for(unsigned int n = sse_iters * 4; n < num_points; n++)
tmp16 = *in_common++; printf("a_sse phase %i: %f,%f\n", n,lv_creal(*phase),lv_cimag(*phase));
tmp32 = lv_cmake((float)lv_creal(tmp16), (float)lv_cimag(tmp16)) * (*phase);
tmp16 = lv_cmake((int16_t)rintf(lv_creal(tmp32)), (int16_t)rintf(lv_cimag(tmp32)));
(*phase) *= phase_inc;
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
{
tmp16 = *in_common++;
tmp32 = lv_cmake((float)lv_creal(tmp16), (float)lv_cimag(tmp16)) * (*phase);
tmp16 = lv_cmake((int16_t)rintf(lv_creal(tmp32)), (int16_t)rintf(lv_cimag(tmp32)));
(*phase) *= phase_inc;
lv_16sc_t tmp = tmp16 * in_a[n_vec][n];
//lv_16sc_t tmp = lv_cmake(sat_adds16i(sat_muls16i(lv_creal(tmp16), lv_creal(in_a[n_vec][n])), - sat_muls16i(lv_cimag(tmp16), lv_cimag(in_a[n_vec][n]))) , sat_adds16i(sat_muls16i(lv_creal(tmp16), lv_cimag(in_a[n_vec][n])), sat_muls16i(lv_cimag(tmp16), lv_creal(in_a[n_vec][n]))));
_out[n_vec] = lv_cmake(sat_adds16i(lv_creal(_out[n_vec]), lv_creal(tmp)),
sat_adds16i(lv_cimag(_out[n_vec]), lv_cimag(tmp)));
}
@ -386,16 +393,17 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_u_sse3(lv_16sc_
free(imagcacc);
_mm_store_ps((float*)two_phase_acc, two_phase_acc_reg);
(*phase) = lv_cmake(two_phase_acc[0], two_phase_acc[0]);
(*phase) = two_phase_acc[0];//lv_cmake(two_phase_acc[0], two_phase_acc[1]);
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
for(unsigned int n = sse_iters * 4; n < num_points; n++)
{
for(unsigned int n = sse_iters * 4; n < num_points; n++)
tmp16 = *in_common++;
tmp32 = lv_cmake((float)lv_creal(tmp16), (float)lv_cimag(tmp16)) * (*phase);
tmp16 = lv_cmake((int16_t)rintf(lv_creal(tmp32)), (int16_t)rintf(lv_cimag(tmp32)));
(*phase) *= phase_inc;
for (int n_vec = 0; n_vec < num_a_vectors; n_vec++)
{
tmp16 = *in_common++;
tmp32 = lv_cmake((float)lv_creal(tmp16), (float)lv_cimag(tmp16)) * (*phase);
tmp16 = lv_cmake((int16_t)rintf(lv_creal(tmp32)), (int16_t)rintf(lv_cimag(tmp32)));
(*phase) *= phase_inc;
lv_16sc_t tmp = tmp16 * in_a[n_vec][n];
_out[n_vec] = lv_cmake(sat_adds16i(lv_creal(_out[n_vec]), lv_creal(tmp)),
sat_adds16i(lv_cimag(_out[n_vec]), lv_cimag(tmp)));
@ -484,7 +492,6 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_neon(lv_16sc_t*
tmp32f.val[1] = vcvtq_f32_s32(tmp32i.val[1]);
/* complex multiplication of four complex samples (float 32 bits each component) */
tmp32_real.val[0] = vmulq_f32(tmp32f.val[0], _phase_real);
tmp32_real.val[1] = vmulq_f32(tmp32f.val[1], _phase_imag);
tmp32_imag.val[0] = vmulq_f32(tmp32f.val[0], _phase_imag);
@ -528,17 +535,17 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_neon(lv_16sc_t*
b_val.val[0] = vmul_s16(a_val.val[0], tmp16.val[0]);
// a0i*b0i|a1i*b1i|a2i*b2i|a3i*b3i
b_val.val[1] = vmul_s16(a_val.val[1], tmp16.val[1]);
c_val.val[0] = vsub_s16(b_val.val[0], b_val.val[1]);
c_val.val[0] = vqsub_s16(b_val.val[0], b_val.val[1]);
// Multiply cross terms to get the imaginary result
// a0r*b0i|a1r*b1i|a2r*b2i|a3r*b3i
b_val.val[0] = vmul_s16(a_val.val[0], tmp16.val[1]);
// a0i*b0r|a1i*b1r|a2i*b2r|a3i*b3r
b_val.val[1] = vmul_s16(a_val.val[1], tmp16.val[0]);
c_val.val[1] = vadd_s16(b_val.val[0], b_val.val[1]);
c_val.val[1] = vqadd_s16(b_val.val[0], b_val.val[1]);
accumulator[n_vec].val[0] = vadd_s16(accumulator[n_vec].val[0], c_val.val[0]);
accumulator[n_vec].val[1] = vadd_s16(accumulator[n_vec].val[1], c_val.val[1]);
accumulator[n_vec].val[0] = vqadd_s16(accumulator[n_vec].val[0], c_val.val[0]);
accumulator[n_vec].val[1] = vqadd_s16(accumulator[n_vec].val[1], c_val.val[1]);
}
}
@ -562,7 +569,7 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn_neon(lv_16sc_t*
for (unsigned int n = neon_iters * 4; n < num_points; n++)
{
tmp16_ = in_common[n];
tmp16_ = in_common[n]; printf("neon phase %i: %f,%f\n", n,lv_creal(*phase),lv_cimag(*phase));
tmp32_ = lv_cmake((float32_t)lv_creal(tmp16_), (float32_t)lv_cimag(tmp16_)) * (*phase);
tmp16_ = lv_cmake((int16_t)rintf(lv_creal(tmp32_)), (int16_t)rintf(lv_cimag(tmp32_)));
(*phase) *= phase_inc;

20
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_rotator_dotprodxnpuppet_16ic.h
View File

@ -37,8 +37,8 @@
#include "volk_gnsssdr/volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn.h"
#include <volk_gnsssdr/volk_gnsssdr_malloc.h>
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
#include <volk_gnsssdr/volk_gnsssdr.h>
//#include <volk_gnsssdr/volk_gnsssdr_complex.h>
//#include <volk_gnsssdr/volk_gnsssdr.h>
#include <string.h>
#ifdef LV_HAVE_GENERIC
@ -46,11 +46,11 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dotprodxnpuppet_16ic_generic(lv_
{
// phases must be normalized. Phase rotator expects a complex exponential input!
float rem_carrier_phase_in_rad = 0.345;
float phase_step_rad = 0.123;
float phase_step_rad = 0.001;
lv_32fc_t phase[1];
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), -sin(rem_carrier_phase_in_rad));
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));
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
int num_a_vectors = 3;
lv_16sc_t** in_a = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
@ -75,11 +75,11 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dotprodxnpuppet_16ic_a_sse3(lv_1
{
// phases must be normalized. Phase rotator expects a complex exponential input!
float rem_carrier_phase_in_rad = 0.345;
float phase_step_rad = 0.123;
float phase_step_rad = 0.001;
lv_32fc_t phase[1];
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), -sin(rem_carrier_phase_in_rad));
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));
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
int num_a_vectors = 3;
lv_16sc_t** in_a = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_a_vectors, volk_gnsssdr_get_alignment());
@ -108,9 +108,9 @@ static inline void volk_gnsssdr_16ic_x2_rotator_dotprodxnpuppet_16ic_u_sse3(lv_1
float rem_carrier_phase_in_rad = 0.345;
float phase_step_rad = 0.123;
lv_32fc_t phase[1];
phase[0] = lv_cmake(cos(rem_carrier_phase_in_rad), -sin(rem_carrier_phase_in_rad));
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));
phase_inc[0] = lv_cmake(cos(phase_step_rad), sin(phase_step_rad));
int num_a_vectors = 3;
lv_16sc_t** in_a = (lv_16sc_t**)volk_gnsssdr_malloc(sizeof(lv_16sc_t*) * num_a_vectors, volk_gnsssdr_get_alignment());

5
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/lib/kernel_tests.h
View File

@ -59,6 +59,9 @@ std::vector<volk_gnsssdr_test_case_t> init_test_list(volk_gnsssdr_test_params_t
volk_gnsssdr_test_params_t test_params_more_iters = volk_gnsssdr_test_params_t(test_params.tol(), test_params.scalar(),
test_params.vlen(), 100000, test_params.benchmark_mode(), test_params.kernel_regex());
volk_gnsssdr_test_params_t test_params_1_iter = volk_gnsssdr_test_params_t(test_params.tol(), test_params.scalar(),
test_params.vlen(), 1, test_params.benchmark_mode(), test_params.kernel_regex());
std::vector<volk_gnsssdr_test_case_t> test_cases = boost::assign::list_of
(VOLK_INIT_TEST(volk_gnsssdr_8i_accumulator_s8i, test_params_more_iters))
@ -81,7 +84,7 @@ std::vector<volk_gnsssdr_test_case_t> init_test_list(volk_gnsssdr_test_params_t
(VOLK_INIT_PUPP(volk_gnsssdr_16ic_resamplerpuppet_16ic, volk_gnsssdr_16ic_resampler_16ic, test_params))
(VOLK_INIT_PUPP(volk_gnsssdr_16ic_resamplerxnpuppet_16ic, volk_gnsssdr_16ic_xn_resampler_16ic_xn, test_params))
(VOLK_INIT_PUPP(volk_gnsssdr_16ic_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))
(VOLK_INIT_PUPP(volk_gnsssdr_16ic_x2_rotator_dotprodxnpuppet_16ic, volk_gnsssdr_16ic_x2_rotator_dot_prod_16ic_xn, test_params_1_iter))
;
return test_cases;