mirrors/gnss-sdr
1
0
Fork 0
mirror of https://github.com/gnss-sdr/gnss-sdr synced 2024-12-14 04:00:34 +00:00
Code Issues Releases Wiki Activity

ask for aligned memory in neon implementation

Browse Source 
1% improvement
This commit is contained in:
Carles Fernandez 2016-01-31 10:39:24 +01:00
parent 4fcffa2bdd
commit 8a6c4d767f
1 changed files with 3 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
src/algorithms/libs/volk_gnsssdr_module/volk_gnsssdr/kernels/volk_gnsssdr/volk_gnsssdr_16ic_x2_dot_prod_16ic.h
View File

@ -77,6 +77,7 @@ static inline void volk_gnsssdr_16ic_x2_dot_prod_16ic_a_sse2(lv_16sc_t* out, con
if (sse_iters > 0) if (sse_iters > 0)
{ {
__m128i a,b,c, c_sr, mask_imag, mask_real, real, imag, imag1,imag2, b_sl, a_sl, realcacc, imagcacc, result; __m128i a,b,c, c_sr, mask_imag, mask_real, real, imag, imag1,imag2, b_sl, a_sl, realcacc, imagcacc, result;
__VOLK_ATTR_ALIGNED(16) lv_16sc_t dotProductVector[4];
realcacc = _mm_setzero_si128(); realcacc = _mm_setzero_si128();
imagcacc = _mm_setzero_si128(); imagcacc = _mm_setzero_si128();
@ -116,8 +117,6 @@ static inline void volk_gnsssdr_16ic_x2_dot_prod_16ic_a_sse2(lv_16sc_t* out, con
result = _mm_or_si128 (realcacc, imagcacc); result = _mm_or_si128 (realcacc, imagcacc);
__VOLK_ATTR_ALIGNED(16) lv_16sc_t dotProductVector[4];
_mm_store_si128((__m128i*)dotProductVector,result); // Store the results back into the dot product vector _mm_store_si128((__m128i*)dotProductVector,result); // Store the results back into the dot product vector
for (int i = 0; i < 4; ++i) for (int i = 0; i < 4; ++i)
@ -154,6 +153,7 @@ static inline void volk_gnsssdr_16ic_x2_dot_prod_16ic_u_sse2(lv_16sc_t* out, con
if (sse_iters > 0) if (sse_iters > 0)
{ {
__m128i a,b,c, c_sr, mask_imag, mask_real, real, imag, imag1,imag2, b_sl, a_sl, realcacc, imagcacc, result; __m128i a,b,c, c_sr, mask_imag, mask_real, real, imag, imag1,imag2, b_sl, a_sl, realcacc, imagcacc, result;
__VOLK_ATTR_ALIGNED(16) lv_16sc_t dotProductVector[4];
realcacc = _mm_setzero_si128(); realcacc = _mm_setzero_si128();
imagcacc = _mm_setzero_si128(); imagcacc = _mm_setzero_si128();
@ -193,8 +193,6 @@ static inline void volk_gnsssdr_16ic_x2_dot_prod_16ic_u_sse2(lv_16sc_t* out, con
result = _mm_or_si128 (realcacc, imagcacc); result = _mm_or_si128 (realcacc, imagcacc);
__VOLK_ATTR_ALIGNED(16) lv_16sc_t dotProductVector[4];
_mm_storeu_si128((__m128i*)dotProductVector,result); // Store the results back into the dot product vector _mm_storeu_si128((__m128i*)dotProductVector,result); // Store the results back into the dot product vector
for (int i = 0; i < 4; ++i) for (int i = 0; i < 4; ++i)
@ -227,7 +225,7 @@ static inline void volk_gnsssdr_16ic_x2_dot_prod_16ic_neon(lv_16sc_t* out, const
// 2nd lane holds the imaginary part // 2nd lane holds the imaginary part
int16x4x2_t a_val, b_val, c_val, accumulator; int16x4x2_t a_val, b_val, c_val, accumulator;
int16x4x2_t tmp_real, tmp_imag; int16x4x2_t tmp_real, tmp_imag;
lv_16sc_t accum_result[4]; __VOLK_ATTR_ALIGNED(16) lv_16sc_t accum_result[4];
accumulator.val[0] = vdup_n_s16(0); accumulator.val[0] = vdup_n_s16(0);
accumulator.val[1] = vdup_n_s16(0); accumulator.val[1] = vdup_n_s16(0);