mirrors/gnss-sdr
1
0
Fork 0
mirror of https://github.com/gnss-sdr/gnss-sdr synced 2024-11-15 22:34:58 +00:00
Code Issues Releases Wiki Activity

some improvements

Browse Source 
phase computation was correclty done in SSE implementation but not in
NEON. Ask for aligned memory in NEON implementation. Some code cleaning
This commit is contained in:
Carles Fernandez 2016-01-31 09:49:50 +01:00
parent db321d1c2e
commit 4fcffa2bdd
1 changed files with 29 additions and 27 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_s32fc_x2_rotator_16ic.h
View File

@ -95,6 +95,8 @@ static inline void volk_gnsssdr_16ic_s32fc_x2_rotator_16ic_a_sse3(lv_16sc_t* out
lv_16sc_t* _out = outVector; lv_16sc_t* _out = outVector;
__m128 yl, yh, tmp1, tmp2, tmp3; __m128 yl, yh, tmp1, tmp2, tmp3;
lv_16sc_t tmp16;
lv_32fc_t tmp32;
for(unsigned int number = 0; number < sse_iters; number++) for(unsigned int number = 0; number < sse_iters; number++)
{ {
@ -146,9 +148,8 @@ static inline void volk_gnsssdr_16ic_s32fc_x2_rotator_16ic_a_sse3(lv_16sc_t* out
} }
_mm_storeu_ps((float*)two_phase_acc, two_phase_acc_reg); _mm_storeu_ps((float*)two_phase_acc, two_phase_acc_reg);
(*phase) = lv_cmake(two_phase_acc[0], two_phase_acc[0]) * phase_inc; (*phase) = lv_cmake(two_phase_acc[0], two_phase_acc[0]);
lv_16sc_t tmp16;
lv_32fc_t tmp32;
for (unsigned int i = sse_iters * 4; i < num_points; ++i) for (unsigned int i = sse_iters * 4; i < num_points; ++i)
{ {
tmp16 = *_in++; tmp16 = *_in++;
@ -182,6 +183,8 @@ static inline void volk_gnsssdr_16ic_s32fc_x2_rotator_16ic_u_sse3(lv_16sc_t* out
lv_16sc_t* _out = outVector; lv_16sc_t* _out = outVector;
__m128 yl, yh, tmp1, tmp2, tmp3; __m128 yl, yh, tmp1, tmp2, tmp3;
lv_16sc_t tmp16;
lv_32fc_t tmp32;
for(unsigned int number = 0; number < sse_iters; number++) for(unsigned int number = 0; number < sse_iters; number++)
{ {
@ -233,9 +236,8 @@ static inline void volk_gnsssdr_16ic_s32fc_x2_rotator_16ic_u_sse3(lv_16sc_t* out
} }
_mm_storeu_ps((float*)two_phase_acc, two_phase_acc_reg); _mm_storeu_ps((float*)two_phase_acc, two_phase_acc_reg);
(*phase) = lv_cmake(two_phase_acc[0], two_phase_acc[0]) * phase_inc; (*phase) = lv_cmake(two_phase_acc[0], two_phase_acc[0]);
lv_16sc_t tmp16;
lv_32fc_t tmp32;
for (unsigned int i = sse_iters * 4; i < num_points; ++i) for (unsigned int i = sse_iters * 4; i < num_points; ++i)
{ {
tmp16 = *_in++; tmp16 = *_in++;
@ -261,8 +263,8 @@ static inline void volk_gnsssdr_16ic_s32fc_x2_rotator_16ic_neon(lv_16sc_t* outVe
lv_16sc_t* _out = outVector; lv_16sc_t* _out = outVector;
lv_32fc_t ___phase4 = phase_inc * phase_inc * phase_inc * phase_inc; lv_32fc_t ___phase4 = phase_inc * phase_inc * phase_inc * phase_inc;
float32_t __phase4_real[4] = { lv_creal(___phase4), lv_creal(___phase4), lv_creal(___phase4), lv_creal(___phase4) }; __VOLK_ATTR_ALIGNED(16) float32_t __phase4_real[4] = { lv_creal(___phase4), lv_creal(___phase4), lv_creal(___phase4), lv_creal(___phase4) };
float32_t __phase4_imag[4] = { lv_cimag(___phase4), lv_cimag(___phase4), lv_cimag(___phase4), lv_cimag(___phase4) }; __VOLK_ATTR_ALIGNED(16) float32_t __phase4_imag[4] = { lv_cimag(___phase4), lv_cimag(___phase4), lv_cimag(___phase4), lv_cimag(___phase4) };
float32x4_t _phase4_real = vld1q_f32(__phase4_real); float32x4_t _phase4_real = vld1q_f32(__phase4_real);
float32x4_t _phase4_imag = vld1q_f32(__phase4_imag); float32x4_t _phase4_imag = vld1q_f32(__phase4_imag);
@ -271,8 +273,8 @@ static inline void volk_gnsssdr_16ic_s32fc_x2_rotator_16ic_neon(lv_16sc_t* outVe
lv_32fc_t phase3 = phase2 * phase_inc; lv_32fc_t phase3 = phase2 * phase_inc;
lv_32fc_t phase4 = phase3 * phase_inc; lv_32fc_t phase4 = phase3 * phase_inc;
float32_t __phase_real[4] = { lv_creal((*phase)), lv_creal(phase2), lv_creal(phase3), lv_creal(phase4) }; __VOLK_ATTR_ALIGNED(16) float32_t __phase_real[4] = { lv_creal((*phase)), lv_creal(phase2), lv_creal(phase3), lv_creal(phase4) };
float32_t __phase_imag[4] = { lv_cimag((*phase)), lv_cimag(phase2), lv_cimag(phase3), lv_cimag(phase4) }; __VOLK_ATTR_ALIGNED(16) float32_t __phase_imag[4] = { lv_cimag((*phase)), lv_cimag(phase2), lv_cimag(phase3), lv_cimag(phase4) };
float32x4_t _phase_real = vld1q_f32(__phase_real); float32x4_t _phase_real = vld1q_f32(__phase_real);
float32x4_t _phase_imag = vld1q_f32(__phase_imag); float32x4_t _phase_imag = vld1q_f32(__phase_imag);
@ -339,7 +341,7 @@ static inline void volk_gnsssdr_16ic_s32fc_x2_rotator_16ic_neon(lv_16sc_t* outVe
vst1q_f32((float32_t*)__phase_real, _phase_real); vst1q_f32((float32_t*)__phase_real, _phase_real);
vst1q_f32((float32_t*)__phase_imag, _phase_imag); vst1q_f32((float32_t*)__phase_imag, _phase_imag);
(*phase) = lv_cmake((float32_t)__phase_real[3], (float32_t)__phase_imag[3]) * phase_inc; (*phase) = lv_cmake((float32_t)__phase_real[0], (float32_t)__phase_imag[0]);
} }
for(i = 0; i < neon_iters % 4; ++i) for(i = 0; i < neon_iters % 4; ++i)
{ {