mirror of
https://github.com/gnss-sdr/gnss-sdr
synced 2024-12-14 12:10:34 +00:00
Fixing resamplers
Under some circumstances (i.e. negative Doppler) it could cause a segmentation fault. It is now fixed for all protokernels.
This commit is contained in:
parent
fc6fdc277c
commit
414eaddb42
@ -64,6 +64,7 @@
|
|||||||
#define INCLUDED_volk_gnsssdr_16ic_xn_resampler2_16ic_xn_H
|
#define INCLUDED_volk_gnsssdr_16ic_xn_resampler2_16ic_xn_H
|
||||||
|
|
||||||
#include <math.h>
|
#include <math.h>
|
||||||
|
#include <stdlib.h>
|
||||||
#include <volk_gnsssdr/volk_gnsssdr_common.h>
|
#include <volk_gnsssdr/volk_gnsssdr_common.h>
|
||||||
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
|
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
|
||||||
|
|
||||||
@ -80,7 +81,7 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_generic(lv_16sc_t** r
|
|||||||
// resample code for current tap
|
// 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 = (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!
|
//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;
|
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;
|
local_code_chip_index = local_code_chip_index % code_length_chips;
|
||||||
result[current_correlator_tap][n] = local_code[local_code_chip_index];
|
result[current_correlator_tap][n] = local_code[local_code_chip_index];
|
||||||
}
|
}
|
||||||
@ -144,7 +145,7 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_a_sse4_1(lv_16sc_t**
|
|||||||
// resample code for current tap
|
// 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_ = (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!
|
//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;
|
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;
|
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
@ -208,7 +209,7 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_u_sse4_1(lv_16sc_t**
|
|||||||
// resample code for current tap
|
// 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_ = (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!
|
//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;
|
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;
|
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
@ -276,7 +277,7 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_a_sse3(lv_16sc_t** re
|
|||||||
// resample code for current tap
|
// 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_ = (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!
|
//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;
|
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;
|
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
@ -344,7 +345,7 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_u_sse3(lv_16sc_t** re
|
|||||||
// resample code for current tap
|
// 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_ = (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!
|
//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;
|
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;
|
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
@ -382,14 +383,13 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_a_avx(lv_16sc_t** res
|
|||||||
indexn = n0;
|
indexn = n0;
|
||||||
for(unsigned int n = 0; n < avx_iters; n++)
|
for(unsigned int n = 0; n < avx_iters; n++)
|
||||||
{
|
{
|
||||||
|
__builtin_prefetch(&_result[current_correlator_tap][8 * n + 7], 1, 0);
|
||||||
|
__builtin_prefetch(&local_code_chip_index[8], 1, 3);
|
||||||
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
|
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
|
||||||
aux = _mm256_add_ps(aux, aux2);
|
aux = _mm256_add_ps(aux, aux2);
|
||||||
// floor
|
// floor
|
||||||
aux = _mm256_floor_ps(aux);
|
aux = _mm256_floor_ps(aux);
|
||||||
|
|
||||||
negatives = _mm256_cmp_ps(aux, zeros, 0x01);
|
|
||||||
aux3 = _mm256_and_ps(code_length_chips_reg_f, negatives);
|
|
||||||
aux = _mm256_add_ps(aux, aux3);
|
|
||||||
// fmod
|
// fmod
|
||||||
c = _mm256_div_ps(aux, code_length_chips_reg_f);
|
c = _mm256_div_ps(aux, code_length_chips_reg_f);
|
||||||
i = _mm256_cvttps_epi32(c);
|
i = _mm256_cvttps_epi32(c);
|
||||||
@ -397,6 +397,13 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_a_avx(lv_16sc_t** res
|
|||||||
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
|
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||||
local_code_chip_index_reg = _mm256_cvttps_epi32(_mm256_sub_ps(aux, base));
|
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);
|
_mm256_store_si256((__m256i*)local_code_chip_index, local_code_chip_index_reg);
|
||||||
for(unsigned int k = 0; k < 8; ++k)
|
for(unsigned int k = 0; k < 8; ++k)
|
||||||
{
|
{
|
||||||
@ -413,7 +420,7 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_a_avx(lv_16sc_t** res
|
|||||||
// resample code for current tap
|
// 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_ = (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!
|
//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;
|
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;
|
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
@ -451,14 +458,13 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_u_avx(lv_16sc_t** res
|
|||||||
indexn = n0;
|
indexn = n0;
|
||||||
for(unsigned int n = 0; n < avx_iters; n++)
|
for(unsigned int n = 0; n < avx_iters; n++)
|
||||||
{
|
{
|
||||||
|
__builtin_prefetch(&_result[current_correlator_tap][8 * n + 7], 1, 0);
|
||||||
|
__builtin_prefetch(&local_code_chip_index[8], 1, 3);
|
||||||
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
|
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
|
||||||
aux = _mm256_add_ps(aux, aux2);
|
aux = _mm256_add_ps(aux, aux2);
|
||||||
// floor
|
// floor
|
||||||
aux = _mm256_floor_ps(aux);
|
aux = _mm256_floor_ps(aux);
|
||||||
|
|
||||||
negatives = _mm256_cmp_ps(aux, zeros, 0x01);
|
|
||||||
aux3 = _mm256_and_ps(code_length_chips_reg_f, negatives);
|
|
||||||
aux = _mm256_add_ps(aux, aux3);
|
|
||||||
// fmod
|
// fmod
|
||||||
c = _mm256_div_ps(aux, code_length_chips_reg_f);
|
c = _mm256_div_ps(aux, code_length_chips_reg_f);
|
||||||
i = _mm256_cvttps_epi32(c);
|
i = _mm256_cvttps_epi32(c);
|
||||||
@ -466,6 +472,13 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_u_avx(lv_16sc_t** res
|
|||||||
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
|
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||||
local_code_chip_index_reg = _mm256_cvttps_epi32(_mm256_sub_ps(aux, base));
|
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);
|
_mm256_store_si256((__m256i*)local_code_chip_index, local_code_chip_index_reg);
|
||||||
for(unsigned int k = 0; k < 8; ++k)
|
for(unsigned int k = 0; k < 8; ++k)
|
||||||
{
|
{
|
||||||
@ -482,7 +495,7 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_u_avx(lv_16sc_t** res
|
|||||||
// resample code for current tap
|
// 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_ = (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!
|
//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;
|
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;
|
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
@ -505,37 +518,39 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_neon(lv_16sc_t** resu
|
|||||||
|
|
||||||
__VOLK_ATTR_ALIGNED(16) int32_t local_code_chip_index[4];
|
__VOLK_ATTR_ALIGNED(16) int32_t local_code_chip_index[4];
|
||||||
int32_t local_code_chip_index_;
|
int32_t local_code_chip_index_;
|
||||||
|
|
||||||
const int32x4_t zeros = vdupq_n_s32(0);
|
const int32x4_t zeros = vdupq_n_s32(0);
|
||||||
const float32x4_t code_length_chips_reg_f = vdupq_n_f32((float)code_length_chips);
|
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);
|
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;
|
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;
|
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 };
|
__VOLK_ATTR_ALIGNED(16) const float vec[4] = { 0.0f, 1.0f, 2.0f, 3.0f };
|
||||||
uint32x4_t igx;
|
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 (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
for (int 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]);
|
shifts_chips_reg = vdupq_n_f32((float)shifts_chips[current_correlator_tap]);
|
||||||
aux2 = vsubq_f32(shifts_chips_reg, rem_code_phase_chips_reg);
|
aux2 = vsubq_f32(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||||
indexn = vld1q_f32((float*)vec);
|
indexn = n0;
|
||||||
for(unsigned int n = 0; n < neon_iters; n++)
|
for(unsigned int n = 0; n < neon_iters; n++)
|
||||||
{
|
{
|
||||||
__builtin_prefetch(&_result[current_correlator_tap][4 * n + 3], 1, 0);
|
__builtin_prefetch(&_result[current_correlator_tap][4 * n + 3], 1, 0);
|
||||||
__builtin_prefetch(&local_code_chip_index[4]);
|
__builtin_prefetch(&local_code_chip_index[4]);
|
||||||
aux = vmulq_f32(code_phase_step_chips_reg, indexn);
|
aux = vmulq_f32(code_phase_step_chips_reg, indexn);
|
||||||
aux = vaddq_f32(aux, aux2);
|
aux = vaddq_f32(aux, aux2);
|
||||||
// floor
|
|
||||||
|
//floor
|
||||||
i = vcvtq_s32_f32(aux);
|
i = vcvtq_s32_f32(aux);
|
||||||
fi = vcvtq_f32_s32(i);
|
fi = vcvtq_f32_s32(i);
|
||||||
igx = vcgtq_f32(fi, aux);
|
igx = vcgtq_f32(fi, aux);
|
||||||
j = vreinterpretq_f32_s32(vandq_s32(vreinterpretq_s32_u32(igx), ones));
|
j = vcvtq_f32_s32(vandq_s32(vreinterpretq_s32_u32(igx), ones));
|
||||||
aux = vsubq_f32(fi, j);
|
aux = vsubq_f32(fi, j);
|
||||||
|
|
||||||
// fmod
|
// fmod
|
||||||
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!
|
|
||||||
c = vmulq_f32(aux, reciprocal);
|
c = vmulq_f32(aux, reciprocal);
|
||||||
i = vcvtq_s32_f32(c);
|
i = vcvtq_s32_f32(c);
|
||||||
cTrunc = vcvtq_f32_s32(i);
|
cTrunc = vcvtq_f32_s32(i);
|
||||||
@ -547,7 +562,8 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_neon(lv_16sc_t** resu
|
|||||||
aux_i = vandq_s32(code_length_chips_reg_i, negatives);
|
aux_i = vandq_s32(code_length_chips_reg_i, negatives);
|
||||||
local_code_chip_index_reg = vaddq_s32(local_code_chip_index_reg, aux_i);
|
local_code_chip_index_reg = vaddq_s32(local_code_chip_index_reg, aux_i);
|
||||||
|
|
||||||
vst1q_s32((int*)local_code_chip_index, local_code_chip_index_reg);
|
vst1q_s32((int32_t*)local_code_chip_index, local_code_chip_index_reg);
|
||||||
|
|
||||||
for(unsigned int k = 0; k < 4; ++k)
|
for(unsigned int k = 0; k < 4; ++k)
|
||||||
{
|
{
|
||||||
_result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
_result[current_correlator_tap][n * 4 + k] = local_code[local_code_chip_index[k]];
|
||||||
@ -558,10 +574,10 @@ static inline void volk_gnsssdr_16ic_xn_resampler2_16ic_xn_neon(lv_16sc_t** resu
|
|||||||
{
|
{
|
||||||
__builtin_prefetch(&_result[current_correlator_tap][n], 1, 0);
|
__builtin_prefetch(&_result[current_correlator_tap][n], 1, 0);
|
||||||
// resample code for current tap
|
// resample code for current tap
|
||||||
local_code_chip_index_ = (int32_t)floor(code_phase_step_chips * (float)n + shifts_chips[current_correlator_tap] - rem_code_phase_chips);
|
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!
|
//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;
|
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_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -64,6 +64,7 @@
|
|||||||
#define INCLUDED_volk_gnsssdr_32fc_xn_resampler_32fc_xn_H
|
#define INCLUDED_volk_gnsssdr_32fc_xn_resampler_32fc_xn_H
|
||||||
|
|
||||||
#include <math.h>
|
#include <math.h>
|
||||||
|
#include <stdlib.h> /* abs */
|
||||||
#include <volk_gnsssdr/volk_gnsssdr_common.h>
|
#include <volk_gnsssdr/volk_gnsssdr_common.h>
|
||||||
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
|
#include <volk_gnsssdr/volk_gnsssdr_complex.h>
|
||||||
|
|
||||||
@ -80,7 +81,7 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_generic(lv_32fc_t** re
|
|||||||
// resample code for current tap
|
// 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 = (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!
|
//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;
|
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;
|
local_code_chip_index = local_code_chip_index % code_length_chips;
|
||||||
result[current_correlator_tap][n] = local_code[local_code_chip_index];
|
result[current_correlator_tap][n] = local_code[local_code_chip_index];
|
||||||
}
|
}
|
||||||
@ -97,8 +98,8 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_sse3(lv_32fc_t** res
|
|||||||
lv_32fc_t** _result = result;
|
lv_32fc_t** _result = result;
|
||||||
const unsigned int quarterPoints = num_points / 4;
|
const unsigned int quarterPoints = num_points / 4;
|
||||||
|
|
||||||
const __m128 ones = _mm_set1_ps(1.);
|
const __m128 ones = _mm_set1_ps(1.0f);
|
||||||
const __m128 fours = _mm_set1_ps(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 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);
|
const __m128 code_phase_step_chips_reg = _mm_set_ps1(code_phase_step_chips);
|
||||||
|
|
||||||
@ -115,7 +116,7 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_sse3(lv_32fc_t** res
|
|||||||
{
|
{
|
||||||
shifts_chips_reg = _mm_set_ps1((float)shifts_chips[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);
|
aux2 = _mm_sub_ps(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||||
__m128 indexn = _mm_set_ps(3., 2., 1., 0.);
|
__m128 indexn = _mm_set_ps(3.0f, 2.0f, 1.0f, 0.0f);
|
||||||
for(unsigned int n = 0; n < quarterPoints; n++)
|
for(unsigned int n = 0; n < quarterPoints; n++)
|
||||||
{
|
{
|
||||||
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
aux = _mm_mul_ps(code_phase_step_chips_reg, indexn);
|
||||||
@ -126,10 +127,9 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_sse3(lv_32fc_t** res
|
|||||||
igx = _mm_cmpgt_ps(fi, aux);
|
igx = _mm_cmpgt_ps(fi, aux);
|
||||||
j = _mm_and_ps(igx, ones);
|
j = _mm_and_ps(igx, ones);
|
||||||
aux = _mm_sub_ps(fi, j);
|
aux = _mm_sub_ps(fi, j);
|
||||||
|
|
||||||
// fmod
|
// fmod
|
||||||
c = _mm_div_ps(aux, code_length_chips_reg_f);
|
c = _mm_div_ps(aux, code_length_chips_reg_f);
|
||||||
i = _mm_cvtps_epi32(c);
|
i = _mm_cvttps_epi32(c);
|
||||||
cTrunc = _mm_cvtepi32_ps(i);
|
cTrunc = _mm_cvtepi32_ps(i);
|
||||||
base = _mm_mul_ps(cTrunc, code_length_chips_reg_f);
|
base = _mm_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||||
local_code_chip_index_reg = _mm_cvtps_epi32(_mm_sub_ps(aux, base));
|
local_code_chip_index_reg = _mm_cvtps_epi32(_mm_sub_ps(aux, base));
|
||||||
@ -149,7 +149,7 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_sse3(lv_32fc_t** res
|
|||||||
// resample code for current tap
|
// 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_ = (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!
|
//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;
|
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;
|
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
@ -217,7 +217,7 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_u_sse3(lv_32fc_t** res
|
|||||||
// resample code for current tap
|
// 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_ = (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!
|
//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;
|
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;
|
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
@ -280,7 +280,7 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_sse4_1(lv_32fc_t** r
|
|||||||
// resample code for current tap
|
// 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_ = (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!
|
//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;
|
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;
|
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
@ -344,7 +344,7 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_u_sse4_1(lv_32fc_t** r
|
|||||||
// resample code for current tap
|
// 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_ = (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!
|
//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;
|
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;
|
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
@ -382,14 +382,13 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_avx(lv_32fc_t** resu
|
|||||||
indexn = n0;
|
indexn = n0;
|
||||||
for(unsigned int n = 0; n < avx_iters; n++)
|
for(unsigned int n = 0; n < avx_iters; n++)
|
||||||
{
|
{
|
||||||
|
__builtin_prefetch(&_result[current_correlator_tap][8 * n + 7], 1, 0);
|
||||||
|
__builtin_prefetch(&local_code_chip_index[8], 1, 3);
|
||||||
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
|
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
|
||||||
aux = _mm256_add_ps(aux, aux2);
|
aux = _mm256_add_ps(aux, aux2);
|
||||||
// floor
|
// floor
|
||||||
aux = _mm256_floor_ps(aux);
|
aux = _mm256_floor_ps(aux);
|
||||||
|
|
||||||
negatives = _mm256_cmp_ps(aux, zeros, 0x01);
|
|
||||||
aux3 = _mm256_and_ps(code_length_chips_reg_f, negatives);
|
|
||||||
aux = _mm256_add_ps(aux, aux3);
|
|
||||||
// fmod
|
// fmod
|
||||||
c = _mm256_div_ps(aux, code_length_chips_reg_f);
|
c = _mm256_div_ps(aux, code_length_chips_reg_f);
|
||||||
i = _mm256_cvttps_epi32(c);
|
i = _mm256_cvttps_epi32(c);
|
||||||
@ -397,6 +396,13 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_avx(lv_32fc_t** resu
|
|||||||
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
|
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||||
local_code_chip_index_reg = _mm256_cvttps_epi32(_mm256_sub_ps(aux, base));
|
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);
|
_mm256_store_si256((__m256i*)local_code_chip_index, local_code_chip_index_reg);
|
||||||
for(unsigned int k = 0; k < 8; ++k)
|
for(unsigned int k = 0; k < 8; ++k)
|
||||||
{
|
{
|
||||||
@ -413,7 +419,7 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_a_avx(lv_32fc_t** resu
|
|||||||
// resample code for current tap
|
// 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_ = (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!
|
//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;
|
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;
|
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
@ -451,14 +457,13 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_u_avx(lv_32fc_t** resu
|
|||||||
indexn = n0;
|
indexn = n0;
|
||||||
for(unsigned int n = 0; n < avx_iters; n++)
|
for(unsigned int n = 0; n < avx_iters; n++)
|
||||||
{
|
{
|
||||||
|
__builtin_prefetch(&_result[current_correlator_tap][8 * n + 7], 1, 0);
|
||||||
|
__builtin_prefetch(&local_code_chip_index[8], 1, 3);
|
||||||
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
|
aux = _mm256_mul_ps(code_phase_step_chips_reg, indexn);
|
||||||
aux = _mm256_add_ps(aux, aux2);
|
aux = _mm256_add_ps(aux, aux2);
|
||||||
// floor
|
// floor
|
||||||
aux = _mm256_floor_ps(aux);
|
aux = _mm256_floor_ps(aux);
|
||||||
|
|
||||||
negatives = _mm256_cmp_ps(aux, zeros, 0x01);
|
|
||||||
aux3 = _mm256_and_ps(code_length_chips_reg_f, negatives);
|
|
||||||
aux = _mm256_add_ps(aux, aux3);
|
|
||||||
// fmod
|
// fmod
|
||||||
c = _mm256_div_ps(aux, code_length_chips_reg_f);
|
c = _mm256_div_ps(aux, code_length_chips_reg_f);
|
||||||
i = _mm256_cvttps_epi32(c);
|
i = _mm256_cvttps_epi32(c);
|
||||||
@ -466,6 +471,13 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_u_avx(lv_32fc_t** resu
|
|||||||
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
|
base = _mm256_mul_ps(cTrunc, code_length_chips_reg_f);
|
||||||
local_code_chip_index_reg = _mm256_cvttps_epi32(_mm256_sub_ps(aux, base));
|
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);
|
_mm256_store_si256((__m256i*)local_code_chip_index, local_code_chip_index_reg);
|
||||||
for(unsigned int k = 0; k < 8; ++k)
|
for(unsigned int k = 0; k < 8; ++k)
|
||||||
{
|
{
|
||||||
@ -482,7 +494,7 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_u_avx(lv_32fc_t** resu
|
|||||||
// resample code for current tap
|
// 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_ = (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!
|
//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;
|
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;
|
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
@ -514,27 +526,31 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_neon(lv_32fc_t** resul
|
|||||||
float32x4_t aux, aux2, shifts_chips_reg, fi, c, j, cTrunc, base, indexn, reciprocal;
|
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 };
|
__VOLK_ATTR_ALIGNED(16) const float vec[4] = { 0.0f, 1.0f, 2.0f, 3.0f };
|
||||||
uint32x4_t igx;
|
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 (int current_correlator_tap = 0; current_correlator_tap < num_out_vectors; current_correlator_tap++)
|
for (int 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]);
|
shifts_chips_reg = vdupq_n_f32((float)shifts_chips[current_correlator_tap]);
|
||||||
aux2 = vsubq_f32(shifts_chips_reg, rem_code_phase_chips_reg);
|
aux2 = vsubq_f32(shifts_chips_reg, rem_code_phase_chips_reg);
|
||||||
indexn = vld1q_f32((float*)vec);
|
indexn = n0;
|
||||||
for(unsigned int n = 0; n < neon_iters; n++)
|
for(unsigned int n = 0; n < neon_iters; n++)
|
||||||
{
|
{
|
||||||
__builtin_prefetch(&_result[current_correlator_tap][4 * n + 3], 1, 0);
|
__builtin_prefetch(&_result[current_correlator_tap][4 * n + 3], 1, 0);
|
||||||
__builtin_prefetch(&local_code_chip_index[4]);
|
__builtin_prefetch(&local_code_chip_index[4]);
|
||||||
aux = vmulq_f32(code_phase_step_chips_reg, indexn);
|
aux = vmulq_f32(code_phase_step_chips_reg, indexn);
|
||||||
aux = vaddq_f32(aux, aux2);
|
aux = vaddq_f32(aux, aux2);
|
||||||
// floor
|
|
||||||
|
//floor
|
||||||
i = vcvtq_s32_f32(aux);
|
i = vcvtq_s32_f32(aux);
|
||||||
fi = vcvtq_f32_s32(i);
|
fi = vcvtq_f32_s32(i);
|
||||||
igx = vcgtq_f32(fi, aux);
|
igx = vcgtq_f32(fi, aux);
|
||||||
j = vcvtq_f32_s32(vandq_s32(vreinterpretq_s32_u32(igx), ones));
|
j = vcvtq_f32_s32(vandq_s32(vreinterpretq_s32_u32(igx), ones));
|
||||||
aux = vsubq_f32(fi, j);
|
aux = vsubq_f32(fi, j);
|
||||||
|
|
||||||
// fmod
|
// fmod
|
||||||
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!
|
|
||||||
c = vmulq_f32(aux, reciprocal);
|
c = vmulq_f32(aux, reciprocal);
|
||||||
i = vcvtq_s32_f32(c);
|
i = vcvtq_s32_f32(c);
|
||||||
cTrunc = vcvtq_f32_s32(i);
|
cTrunc = vcvtq_f32_s32(i);
|
||||||
@ -560,7 +576,7 @@ static inline void volk_gnsssdr_32fc_xn_resampler_32fc_xn_neon(lv_32fc_t** resul
|
|||||||
// resample code for current tap
|
// 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_ = (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!
|
//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;
|
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;
|
local_code_chip_index_ = local_code_chip_index_ % code_length_chips;
|
||||||
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
_result[current_correlator_tap][n] = local_code[local_code_chip_index_];
|
||||||
}
|
}
|
||||||
|
Loading…
Reference in New Issue
Block a user