Added ILP after AVX-512
This commit is contained in:
@@ -1,7 +1,6 @@
|
||||
// # Changes:
|
||||
// # - Ported AVX2 to AVX-512
|
||||
// # - Replaced Obstacle with the version that contains no Vector2d
|
||||
// # - 64-byte aligned the vectors + replaced loadu with load
|
||||
// # - Implemented Fast Inverse Square Root
|
||||
// # Predecessor: variant_inlining_7_SoA_AVX2_polyexp.cpp
|
||||
|
||||
@@ -292,11 +291,11 @@ int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, c
|
||||
__mmask8 m_in_range = _mm512_cmp_pd_mask(j_idx, v_np, _CMP_LT_OQ);
|
||||
__mmask8 active = m_neq_i & m_in_range;
|
||||
|
||||
__m512d pjx = _mm512_load_pd(&pos_x[j]);
|
||||
__m512d pjy = _mm512_load_pd(&pos_y[j]);
|
||||
__m512d sbe_jx = _mm512_load_pd(&sb_e_x[j]);
|
||||
__m512d sbe_jy = _mm512_load_pd(&sb_e_y[j]);
|
||||
__m512d sb_jsq = _mm512_load_pd(&sb_sq[j]);
|
||||
__m512d pjx = _mm512_loadu_pd(&pos_x[j]);
|
||||
__m512d pjy = _mm512_loadu_pd(&pos_y[j]);
|
||||
__m512d sbe_jx = _mm512_loadu_pd(&sb_e_x[j]);
|
||||
__m512d sbe_jy = _mm512_loadu_pd(&sb_e_y[j]);
|
||||
__m512d sb_jsq = _mm512_loadu_pd(&sb_sq[j]);
|
||||
|
||||
__m512d r_ab_x = _mm512_sub_pd(pix, pjx);
|
||||
__m512d r_ab_y = _mm512_sub_pd(piy, pjy);
|
||||
|
||||
Reference in New Issue
Block a user