From a7d45665dabb8d795a8dccabc241af7a8c0271b2 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=5BQuy=20Anh=5D=20=C2=ABElliot=C2=BB=20Nguyen?=
 <23906-nguyenquy@users.noreply.gitlab.inf.ethz.ch>
Date: Wed, 17 Jun 2026 07:28:11 +0200
Subject: [PATCH] Flop Count Calculations + x86 Measurements + c13 Fix

---
 headers/constants.hpp                         |  15 +
 headers/constants_float.hpp                   |  27 +-
 measurements/memory/x86/1.log                 |  24 +-
 measurements/memory/x86/10.log                |  37 +
 measurements/memory/x86/100.log               |  37 +
 measurements/memory/x86/11.log                |  37 +
 measurements/memory/x86/12.log                |  37 +
 measurements/memory/x86/13.log                |  37 +
 measurements/memory/x86/14.log                |  37 +
 measurements/memory/x86/15.log                |  37 +
 measurements/memory/x86/16.log                |  37 +
 measurements/memory/x86/17.log                |  37 +
 measurements/memory/x86/18.log                |  37 +
 measurements/memory/x86/19.log                |  37 +
 measurements/memory/x86/2.log                 |  24 +-
 measurements/memory/x86/20.log                |  37 +
 measurements/memory/x86/21.log                |  37 +
 measurements/memory/x86/22.log                |  37 +
 measurements/memory/x86/23.log                |  37 +
 measurements/memory/x86/24.log                |  37 +
 measurements/memory/x86/25.log                |  37 +
 measurements/memory/x86/26.log                |  37 +
 measurements/memory/x86/27.log                |  37 +
 measurements/memory/x86/28.log                |  37 +
 measurements/memory/x86/29.log                |  37 +
 measurements/memory/x86/3.log                 |  24 +-
 measurements/memory/x86/30.log                |  37 +
 measurements/memory/x86/31.log                |  37 +
 measurements/memory/x86/32.log                |  37 +
 measurements/memory/x86/33.log                |  37 +
 measurements/memory/x86/34.log                |  37 +
 measurements/memory/x86/35.log                |  37 +
 measurements/memory/x86/36.log                |  37 +
 measurements/memory/x86/37.log                |  37 +
 measurements/memory/x86/38.log                |  37 +
 measurements/memory/x86/39.log                |  37 +
 measurements/memory/x86/4.log                 |  37 +
 measurements/memory/x86/40.log                |  37 +
 measurements/memory/x86/41.log                |  37 +
 measurements/memory/x86/42.log                |  37 +
 measurements/memory/x86/43.log                |  37 +
 measurements/memory/x86/44.log                |  37 +
 measurements/memory/x86/45.log                |  37 +
 measurements/memory/x86/46.log                |  37 +
 measurements/memory/x86/47.log                |  37 +
 measurements/memory/x86/48.log                |  37 +
 measurements/memory/x86/49.log                |  37 +
 measurements/memory/x86/5.log                 |  37 +
 measurements/memory/x86/50.log                |  37 +
 measurements/memory/x86/51.log                |  37 +
 measurements/memory/x86/52.log                |  37 +
 measurements/memory/x86/53.log                |  37 +
 measurements/memory/x86/54.log                |  37 +
 measurements/memory/x86/55.log                |  37 +
 measurements/memory/x86/56.log                |  37 +
 measurements/memory/x86/57.log                |  37 +
 measurements/memory/x86/58.log                |  37 +
 measurements/memory/x86/59.log                |  37 +
 measurements/memory/x86/6.log                 |  37 +
 measurements/memory/x86/60.log                |  37 +
 measurements/memory/x86/61.log                |  37 +
 measurements/memory/x86/62.log                |  37 +
 measurements/memory/x86/63.log                |  37 +
 measurements/memory/x86/64.log                |  37 +
 measurements/memory/x86/65.log                |  37 +
 measurements/memory/x86/66.log                |  37 +
 measurements/memory/x86/67.log                |  37 +
 measurements/memory/x86/68.log                |  37 +
 measurements/memory/x86/69.log                |  37 +
 measurements/memory/x86/7.log                 |  37 +
 measurements/memory/x86/70.log                |  37 +
 measurements/memory/x86/71.log                |  37 +
 measurements/memory/x86/72.log                |  37 +
 measurements/memory/x86/73.log                |  37 +
 measurements/memory/x86/74.log                |  37 +
 measurements/memory/x86/75.log                |  37 +
 measurements/memory/x86/76.log                |  37 +
 measurements/memory/x86/77.log                |  37 +
 measurements/memory/x86/78.log                |  37 +
 measurements/memory/x86/79.log                |  37 +
 measurements/memory/x86/8.log                 |  37 +
 measurements/memory/x86/80.log                |  37 +
 measurements/memory/x86/81.log                |  37 +
 measurements/memory/x86/82.log                |  37 +
 measurements/memory/x86/83.log                |  37 +
 measurements/memory/x86/84.log                |  37 +
 measurements/memory/x86/85.log                |  37 +
 measurements/memory/x86/86.log                |  37 +
 measurements/memory/x86/87.log                |  37 +
 measurements/memory/x86/88.log                |  37 +
 measurements/memory/x86/89.log                |  37 +
 measurements/memory/x86/9.log                 |  37 +
 measurements/memory/x86/90.log                |  37 +
 measurements/memory/x86/91.log                |  37 +
 measurements/memory/x86/92.log                |  37 +
 measurements/memory/x86/93.log                |  37 +
 measurements/memory/x86/94.log                |  37 +
 measurements/memory/x86/95.log                |  37 +
 measurements/memory/x86/96.log                |  37 +
 measurements/memory/x86/97.log                |  37 +
 measurements/memory/x86/98.log                |  37 +
 measurements/memory/x86/99.log                |  37 +
 measurements/variant_baseline/x86/3.log       |  58 ++
 measurements/variant_branchless/x86/2.log     |  58 ++
 measurements/variant_branchless/x86/3.log     |  58 ++
 .../variant_branchless_sqrtfree/x86/2.log     |  58 ++
 .../variant_branchless_sqrtfree/x86/3.log     |  58 ++
 measurements/variant_cutoff/x86/2.log         |  58 ++
 measurements/variant_cutoff/x86/3.log         |  58 ++
 .../variant_cutoff_callsite/x86/2.log         |  58 ++
 .../variant_cutoff_callsite/x86/3.log         |  58 ++
 measurements/variant_float_baseline/x86/2.log |  58 ++
 measurements/variant_float_baseline/x86/3.log |  58 ++
 measurements/variant_float_inlining/x86/2.log |  58 ++
 measurements/variant_float_inlining/x86/3.log |  58 ++
 .../variant_float_inlining_SoA/x86/2.log      |  58 ++
 .../variant_float_inlining_SoA/x86/3.log      |  58 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  66 ++
 .../x86/3.log                                 |  64 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  64 ++
 .../x86/3.log                                 |  64 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  64 ++
 .../x86/3.log                                 |  64 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  66 ++
 .../x86/3.log                                 |  66 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  66 ++
 .../x86/3.log                                 |  66 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  66 ++
 .../x86/3.log                                 |  66 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  66 ++
 .../x86/3.log                                 |  66 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  66 ++
 .../x86/3.log                                 |  66 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  66 ++
 .../x86/3.log                                 |  66 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  66 ++
 .../x86/3.log                                 |  66 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  64 ++
 .../x86/3.log                                 |  64 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  64 ++
 .../x86/3.log                                 |  64 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  66 ++
 .../x86/3.log                                 |  66 ++
 .../x86/2.log                                 |  58 ++
 .../x86/3.log                                 |  58 ++
 .../x86/2.log                                 |  58 ++
 .../x86/3.log                                 |  58 ++
 .../x86/2.log                                 |  58 ++
 .../x86/3.log                                 |  58 ++
 .../x86/2.log                                 |  58 ++
 .../x86/3.log                                 |  58 ++
 .../x86/2.log                                 |  58 ++
 .../x86/3.log                                 |  58 ++
 measurements/variant_inlining_7/x86/2.log     |  58 ++
 measurements/variant_inlining_7/x86/3.log     |  58 ++
 measurements/variant_inlining_7_SoA/x86/2.log |  58 ++
 measurements/variant_inlining_7_SoA/x86/3.log |  58 ++
 .../variant_inlining_7_SoA_AVX2/x86/3.log     |  66 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  64 ++
 .../x86/3.log                                 |  64 ++
 .../variant_inlining_7_symmetry_1/x86/2.log   |  58 ++
 .../variant_inlining_7_symmetry_1/x86/3.log   |  58 ++
 .../variant_inlining_7_symmetry_2/x86/2.log   |  58 ++
 .../variant_inlining_7_symmetry_2/x86/3.log   |  58 ++
 .../x86/2.log                                 |  58 ++
 .../x86/3.log                                 |  58 ++
 .../x86/2.log                                 |  58 ++
 .../x86/3.log                                 |  58 ++
 measurements/variant_inlining_8/x86/2.log     |  58 ++
 measurements/variant_inlining_8/x86/3.log     |  58 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  66 ++
 .../x86/3.log                                 |  66 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  66 ++
 .../x86/3.log                                 |  66 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  66 ++
 .../x86/3.log                                 |  66 ++
 .../x86/1.log                                 |  66 ++
 .../x86/2.log                                 |  66 ++
 .../x86/3.log                                 |  66 ++
 measurements/variant_lut_exp/x86/2.log        |  58 ++
 measurements/variant_lut_exp/x86/3.log        |  58 ++
 scripts/mem-bench.sh                          |  12 +-
 variants/variant_AVX512_1_flopcount.cpp       | 568 ++++++++++++++
 variants/variant_baseline_flopcount.cpp       | 440 +++++++++++
 variants/variant_float_baseline_flopcount.cpp | 438 +++++++++++
 ...float_inlining_SoA_AVX2_polyexp_expfix.cpp |   3 +-
 ...inlining_SoA_AVX2_polyexp_rsqrt_expfix.cpp |   3 +-
 ..._SoA_AVX2_polyexp_rsqrt_loopopt_expfix.cpp |   3 +-
 ...oat_inlining_SoA_AVX512_polyexp_expfix.cpp | 540 +++++++++++++
 ..._SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp | 557 +++++++++++++
 ...ning_SoA_AVX512_polyexp_moreavx_expfix.cpp | 577 ++++++++++++++
 ...oA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp | 638 +++++++++++++++
 ...2_polyexp_moreavx_ilp_2_moreilp_expfix.cpp | 735 ++++++++++++++++++
 ...512_polyexp_moreavx_ilp_2_rsqrt_expfix.cpp | 634 +++++++++++++++
 ...p_moreavx_ilp_2_rsqrt_morersqrt_expfix.cpp | 206 ++---
 ..._ilp_2_rsqrt_morersqrt_expfix_improved.cpp |   3 +-
 ...sqrt_morersqrt_expfix_improved_moreilp.cpp |   3 +-
 ...oA_AVX512_polyexp_moreavx_ilp_4_expfix.cpp | 727 +++++++++++++++++
 .../variant_float_inlining_SoA_flopcount.cpp  | 380 +++++++++
 variants/variant_float_inlining_flopcount.cpp | 366 +++++++++
 ...ant_inlining_7_SoA_AVX2_polyexp_expfix.cpp |  30 +-
 variants/variant_inlining_7_SoA_flopcount.cpp | 362 +++++++++
 variants/variant_inlining_7_flopcount.cpp     | 352 +++++++++
 ..._SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp | 483 ++++++++++++
 ...A_AVX512_polyexp_fastinvsqrt_sr_expfix.cpp | 576 ++++++++++++++
 ...ng_8_SoA_AVX512_polyexp_moreavx_expfix.cpp | 563 ++++++++++++++
 ...oA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp | 640 +++++++++++++++
 variants/variant_inlining_8_flopcount.cpp     | 343 ++++++++
 225 files changed, 19699 insertions(+), 172 deletions(-)
 create mode 100644 measurements/memory/x86/10.log
 create mode 100644 measurements/memory/x86/100.log
 create mode 100644 measurements/memory/x86/11.log
 create mode 100644 measurements/memory/x86/12.log
 create mode 100644 measurements/memory/x86/13.log
 create mode 100644 measurements/memory/x86/14.log
 create mode 100644 measurements/memory/x86/15.log
 create mode 100644 measurements/memory/x86/16.log
 create mode 100644 measurements/memory/x86/17.log
 create mode 100644 measurements/memory/x86/18.log
 create mode 100644 measurements/memory/x86/19.log
 create mode 100644 measurements/memory/x86/20.log
 create mode 100644 measurements/memory/x86/21.log
 create mode 100644 measurements/memory/x86/22.log
 create mode 100644 measurements/memory/x86/23.log
 create mode 100644 measurements/memory/x86/24.log
 create mode 100644 measurements/memory/x86/25.log
 create mode 100644 measurements/memory/x86/26.log
 create mode 100644 measurements/memory/x86/27.log
 create mode 100644 measurements/memory/x86/28.log
 create mode 100644 measurements/memory/x86/29.log
 create mode 100644 measurements/memory/x86/30.log
 create mode 100644 measurements/memory/x86/31.log
 create mode 100644 measurements/memory/x86/32.log
 create mode 100644 measurements/memory/x86/33.log
 create mode 100644 measurements/memory/x86/34.log
 create mode 100644 measurements/memory/x86/35.log
 create mode 100644 measurements/memory/x86/36.log
 create mode 100644 measurements/memory/x86/37.log
 create mode 100644 measurements/memory/x86/38.log
 create mode 100644 measurements/memory/x86/39.log
 create mode 100644 measurements/memory/x86/4.log
 create mode 100644 measurements/memory/x86/40.log
 create mode 100644 measurements/memory/x86/41.log
 create mode 100644 measurements/memory/x86/42.log
 create mode 100644 measurements/memory/x86/43.log
 create mode 100644 measurements/memory/x86/44.log
 create mode 100644 measurements/memory/x86/45.log
 create mode 100644 measurements/memory/x86/46.log
 create mode 100644 measurements/memory/x86/47.log
 create mode 100644 measurements/memory/x86/48.log
 create mode 100644 measurements/memory/x86/49.log
 create mode 100644 measurements/memory/x86/5.log
 create mode 100644 measurements/memory/x86/50.log
 create mode 100644 measurements/memory/x86/51.log
 create mode 100644 measurements/memory/x86/52.log
 create mode 100644 measurements/memory/x86/53.log
 create mode 100644 measurements/memory/x86/54.log
 create mode 100644 measurements/memory/x86/55.log
 create mode 100644 measurements/memory/x86/56.log
 create mode 100644 measurements/memory/x86/57.log
 create mode 100644 measurements/memory/x86/58.log
 create mode 100644 measurements/memory/x86/59.log
 create mode 100644 measurements/memory/x86/6.log
 create mode 100644 measurements/memory/x86/60.log
 create mode 100644 measurements/memory/x86/61.log
 create mode 100644 measurements/memory/x86/62.log
 create mode 100644 measurements/memory/x86/63.log
 create mode 100644 measurements/memory/x86/64.log
 create mode 100644 measurements/memory/x86/65.log
 create mode 100644 measurements/memory/x86/66.log
 create mode 100644 measurements/memory/x86/67.log
 create mode 100644 measurements/memory/x86/68.log
 create mode 100644 measurements/memory/x86/69.log
 create mode 100644 measurements/memory/x86/7.log
 create mode 100644 measurements/memory/x86/70.log
 create mode 100644 measurements/memory/x86/71.log
 create mode 100644 measurements/memory/x86/72.log
 create mode 100644 measurements/memory/x86/73.log
 create mode 100644 measurements/memory/x86/74.log
 create mode 100644 measurements/memory/x86/75.log
 create mode 100644 measurements/memory/x86/76.log
 create mode 100644 measurements/memory/x86/77.log
 create mode 100644 measurements/memory/x86/78.log
 create mode 100644 measurements/memory/x86/79.log
 create mode 100644 measurements/memory/x86/8.log
 create mode 100644 measurements/memory/x86/80.log
 create mode 100644 measurements/memory/x86/81.log
 create mode 100644 measurements/memory/x86/82.log
 create mode 100644 measurements/memory/x86/83.log
 create mode 100644 measurements/memory/x86/84.log
 create mode 100644 measurements/memory/x86/85.log
 create mode 100644 measurements/memory/x86/86.log
 create mode 100644 measurements/memory/x86/87.log
 create mode 100644 measurements/memory/x86/88.log
 create mode 100644 measurements/memory/x86/89.log
 create mode 100644 measurements/memory/x86/9.log
 create mode 100644 measurements/memory/x86/90.log
 create mode 100644 measurements/memory/x86/91.log
 create mode 100644 measurements/memory/x86/92.log
 create mode 100644 measurements/memory/x86/93.log
 create mode 100644 measurements/memory/x86/94.log
 create mode 100644 measurements/memory/x86/95.log
 create mode 100644 measurements/memory/x86/96.log
 create mode 100644 measurements/memory/x86/97.log
 create mode 100644 measurements/memory/x86/98.log
 create mode 100644 measurements/memory/x86/99.log
 create mode 100644 measurements/variant_baseline/x86/3.log
 create mode 100644 measurements/variant_branchless/x86/2.log
 create mode 100644 measurements/variant_branchless/x86/3.log
 create mode 100644 measurements/variant_branchless_sqrtfree/x86/2.log
 create mode 100644 measurements/variant_branchless_sqrtfree/x86/3.log
 create mode 100644 measurements/variant_cutoff/x86/2.log
 create mode 100644 measurements/variant_cutoff/x86/3.log
 create mode 100644 measurements/variant_cutoff_callsite/x86/2.log
 create mode 100644 measurements/variant_cutoff_callsite/x86/3.log
 create mode 100644 measurements/variant_float_baseline/x86/2.log
 create mode 100644 measurements/variant_float_baseline/x86/3.log
 create mode 100644 measurements/variant_float_inlining/x86/2.log
 create mode 100644 measurements/variant_float_inlining/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX2_polyexp_expfix/x86/1.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX2_polyexp_expfix/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX2_polyexp_expfix/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix/x86/1.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix/x86/1.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_expfix/x86/1.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_expfix/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_expfix/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/1.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix/x86/1.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/1.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix/x86/1.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix/x86/1.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix/x86/1.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved/x86/1.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp/x86/1.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix/x86/1.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_symmetry/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_symmetry/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_symmetry_ILP_16/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_symmetry_ILP_16/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_symmetry_ILP_2/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_symmetry_ILP_2/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_symmetry_ILP_4/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_symmetry_ILP_4/x86/3.log
 create mode 100644 measurements/variant_float_inlining_SoA_symmetry_ILP_8/x86/2.log
 create mode 100644 measurements/variant_float_inlining_SoA_symmetry_ILP_8/x86/3.log
 create mode 100644 measurements/variant_inlining_7/x86/2.log
 create mode 100644 measurements/variant_inlining_7/x86/3.log
 create mode 100644 measurements/variant_inlining_7_SoA/x86/2.log
 create mode 100644 measurements/variant_inlining_7_SoA/x86/3.log
 create mode 100644 measurements/variant_inlining_7_SoA_AVX2/x86/3.log
 create mode 100644 measurements/variant_inlining_7_SoA_AVX2_polyexp_expfix/x86/1.log
 create mode 100644 measurements/variant_inlining_7_SoA_AVX2_polyexp_expfix/x86/2.log
 create mode 100644 measurements/variant_inlining_7_SoA_AVX2_polyexp_expfix/x86/3.log
 create mode 100644 measurements/variant_inlining_7_symmetry_1/x86/2.log
 create mode 100644 measurements/variant_inlining_7_symmetry_1/x86/3.log
 create mode 100644 measurements/variant_inlining_7_symmetry_2/x86/2.log
 create mode 100644 measurements/variant_inlining_7_symmetry_2/x86/3.log
 create mode 100644 measurements/variant_inlining_7_symmetry_2_ilp_1/x86/2.log
 create mode 100644 measurements/variant_inlining_7_symmetry_2_ilp_1/x86/3.log
 create mode 100644 measurements/variant_inlining_7_symmetry_2_ilp_2/x86/2.log
 create mode 100644 measurements/variant_inlining_7_symmetry_2_ilp_2/x86/3.log
 create mode 100644 measurements/variant_inlining_8/x86/2.log
 create mode 100644 measurements/variant_inlining_8/x86/3.log
 create mode 100644 measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/1.log
 create mode 100644 measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/2.log
 create mode 100644 measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/3.log
 create mode 100644 measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix/x86/1.log
 create mode 100644 measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix/x86/2.log
 create mode 100644 measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix/x86/3.log
 create mode 100644 measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix/x86/1.log
 create mode 100644 measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix/x86/2.log
 create mode 100644 measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix/x86/3.log
 create mode 100644 measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/1.log
 create mode 100644 measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/2.log
 create mode 100644 measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/3.log
 create mode 100644 measurements/variant_lut_exp/x86/2.log
 create mode 100644 measurements/variant_lut_exp/x86/3.log
 create mode 100644 variants/variant_AVX512_1_flopcount.cpp
 create mode 100644 variants/variant_baseline_flopcount.cpp
 create mode 100644 variants/variant_float_baseline_flopcount.cpp
 create mode 100644 variants/variant_float_inlining_SoA_AVX512_polyexp_expfix.cpp
 create mode 100644 variants/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp
 create mode 100644 variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix.cpp
 create mode 100644 variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp
 create mode 100644 variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix.cpp
 create mode 100644 variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix.cpp
 create mode 100644 variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix.cpp
 create mode 100644 variants/variant_float_inlining_SoA_flopcount.cpp
 create mode 100644 variants/variant_float_inlining_flopcount.cpp
 create mode 100644 variants/variant_inlining_7_SoA_flopcount.cpp
 create mode 100644 variants/variant_inlining_7_flopcount.cpp
 create mode 100644 variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp
 create mode 100644 variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix.cpp
 create mode 100644 variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix.cpp
 create mode 100644 variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp
 create mode 100644 variants/variant_inlining_8_flopcount.cpp

diff --git a/headers/constants.hpp b/headers/constants.hpp
index ce3c51f..cc7c343 100644
--- a/headers/constants.hpp
+++ b/headers/constants.hpp
@@ -117,3 +117,18 @@ const double CUTOFF_SQ = CUTOFF * CUTOFF;
 const double cos_phi_sq = cos_phi * cos_phi;
 
 const size_t CACHE_BLOCK_SIZE = std::hardware_destructive_interference_size;
+
+const double c0_scalar = 1.0 / 1;
+const double c1_scalar = 1.0 / 1;
+const double c2_scalar = 1.0 / 2;
+const double c3_scalar = 1.0 / 6;
+const double c4_scalar = 1.0 / 24;
+const double c5_scalar = 1.0 / 120;
+const double c6_scalar = 1.0 / 720;
+const double c7_scalar = 1.0 / 5040;
+const double c8_scalar = 1.0 / 40320;
+const double c9_scalar = 1.0 / 362880;
+const double c10_scalar = 1.0 / 3628800;
+const double c11_scalar = 1.0 / 39916800;
+const double c12_scalar = 1.0 / 479001600;
+const double c13_scalar = 1.0 / 6227020800;
diff --git a/headers/constants_float.hpp b/headers/constants_float.hpp
index 4f42c62..c8c2126 100644
--- a/headers/constants_float.hpp
+++ b/headers/constants_float.hpp
@@ -104,16 +104,17 @@ const float cos_phi_sq = cos_phi * cos_phi;
 
 const size_t CACHE_BLOCK_SIZE = std::hardware_destructive_interference_size;
 
-const double c0_scalar = 1.0 / 1;
-const double c1_scalar = 1.0 / 1;
-const double c2_scalar = 1.0 / 2;
-const double c3_scalar = 1.0 / 6;
-const double c4_scalar = 1.0 / 24;
-const double c5_scalar = 1.0 / 120;
-const double c6_scalar = 1.0 / 720;
-const double c7_scalar = 1.0 / 5040;
-const double c8_scalar = 1.0 / 40320;
-const double c9_scalar = 1.0 / 362880;
-const double c10_scalar = 1.0 / 3628800;
-const double c11_scalar = 1.0 / 39916800;
-const double c12_scalar = 1.0 / 479001600;
+const float c0_scalar = 1.0 / 1;
+const float c1_scalar = 1.0 / 1;
+const float c2_scalar = 1.0 / 2;
+const float c3_scalar = 1.0 / 6;
+const float c4_scalar = 1.0 / 24;
+const float c5_scalar = 1.0 / 120;
+const float c6_scalar = 1.0 / 720;
+const float c7_scalar = 1.0 / 5040;
+const float c8_scalar = 1.0 / 40320;
+const float c9_scalar = 1.0 / 362880;
+const float c10_scalar = 1.0 / 3628800;
+const float c11_scalar = 1.0 / 39916800;
+const float c12_scalar = 1.0 / 479001600;
+const float c13_scalar = 1.0 / 6227020800;
diff --git a/measurements/memory/x86/1.log b/measurements/memory/x86/1.log
index 76a392f..20be596 100644
--- a/measurements/memory/x86/1.log
+++ b/measurements/memory/x86/1.log
@@ -1,7 +1,7 @@
 sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
 
 Running the test with following options:
-Number of threads: 1
+Number of threads: 8
 Initializing random number generator from current time
 
 
@@ -15,23 +15,23 @@ Initializing worker threads...
 
 Threads started!
 
-Total operations: 20 (   37.08 per second)
+Total operations: 16 (   94.31 per second)
 
-20480.00 MiB transferred (37967.45 MiB/sec)
+16384.00 MiB transferred (96577.44 MiB/sec)
 
 
 General statistics:
-    total time:                          0.5378s
-    total number of events:              20
+    total time:                          0.1680s
+    total number of events:              16
 
 Latency (ms):
-         min:                                   25.99
-         avg:                                   26.87
-         max:                                   33.18
-         95th percentile:                       28.16
-         sum:                                  537.47
+         min:                                   73.61
+         avg:                                   81.30
+         max:                                   94.24
+         95th percentile:                       90.78
+         sum:                                 1300.79
 
 Threads fairness:
-    events (avg/stddev):           20.0000/0.00
-    execution time (avg/stddev):   0.5375/0.00
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1626/0.00
 
diff --git a/measurements/memory/x86/10.log b/measurements/memory/x86/10.log
new file mode 100644
index 0000000..e1355c5
--- /dev/null
+++ b/measurements/memory/x86/10.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   79.70 per second)
+
+16384.00 MiB transferred (81607.97 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1990s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   73.93
+         avg:                                   95.91
+         max:                                  115.12
+         95th percentile:                      114.72
+         sum:                                 1534.52
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1918/0.01
+
diff --git a/measurements/memory/x86/100.log b/measurements/memory/x86/100.log
new file mode 100644
index 0000000..af704aa
--- /dev/null
+++ b/measurements/memory/x86/100.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   66.20 per second)
+
+16384.00 MiB transferred (67785.69 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2400s
+    total number of events:              16
+
+Latency (ms):
+         min:                                  103.07
+         avg:                                  118.76
+         max:                                  133.11
+         95th percentile:                      132.49
+         sum:                                 1900.15
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2375/0.00
+
diff --git a/measurements/memory/x86/11.log b/measurements/memory/x86/11.log
new file mode 100644
index 0000000..64e6da9
--- /dev/null
+++ b/measurements/memory/x86/11.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   65.88 per second)
+
+16384.00 MiB transferred (67456.51 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2412s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   66.60
+         avg:                                  106.05
+         max:                                  174.46
+         95th percentile:                      155.80
+         sum:                                 1696.87
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2121/0.03
+
diff --git a/measurements/memory/x86/12.log b/measurements/memory/x86/12.log
new file mode 100644
index 0000000..f3e747a
--- /dev/null
+++ b/measurements/memory/x86/12.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   66.47 per second)
+
+16384.00 MiB transferred (68069.01 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2390s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   53.56
+         avg:                                  101.79
+         max:                                  185.35
+         95th percentile:                      158.63
+         sum:                                 1628.59
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2036/0.03
+
diff --git a/measurements/memory/x86/13.log b/measurements/memory/x86/13.log
new file mode 100644
index 0000000..59104ff
--- /dev/null
+++ b/measurements/memory/x86/13.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   77.60 per second)
+
+16384.00 MiB transferred (79462.84 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2045s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   43.46
+         avg:                                   88.02
+         max:                                  160.68
+         95th percentile:                      153.02
+         sum:                                 1408.36
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1760/0.02
+
diff --git a/measurements/memory/x86/14.log b/measurements/memory/x86/14.log
new file mode 100644
index 0000000..3263ba3
--- /dev/null
+++ b/measurements/memory/x86/14.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   85.35 per second)
+
+16384.00 MiB transferred (87397.03 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1858s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   62.81
+         avg:                                   83.43
+         max:                                  122.67
+         95th percentile:                      123.28
+         sum:                                 1334.89
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1669/0.01
+
diff --git a/measurements/memory/x86/15.log b/measurements/memory/x86/15.log
new file mode 100644
index 0000000..94633c9
--- /dev/null
+++ b/measurements/memory/x86/15.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   79.07 per second)
+
+16384.00 MiB transferred (80971.56 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2007s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   41.20
+         avg:                                   85.98
+         max:                                  159.24
+         95th percentile:                      137.35
+         sum:                                 1375.63
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1720/0.02
+
diff --git a/measurements/memory/x86/16.log b/measurements/memory/x86/16.log
new file mode 100644
index 0000000..6ba2d6c
--- /dev/null
+++ b/measurements/memory/x86/16.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  106.63 per second)
+
+16384.00 MiB transferred (109191.89 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1484s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   68.17
+         avg:                                   71.87
+         max:                                   76.14
+         95th percentile:                       75.82
+         sum:                                 1149.93
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1437/0.00
+
diff --git a/measurements/memory/x86/17.log b/measurements/memory/x86/17.log
new file mode 100644
index 0000000..6444edd
--- /dev/null
+++ b/measurements/memory/x86/17.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  107.19 per second)
+
+16384.00 MiB transferred (109757.98 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1476s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   71.89
+         avg:                                   72.74
+         max:                                   74.50
+         95th percentile:                       74.46
+         sum:                                 1163.80
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1455/0.00
+
diff --git a/measurements/memory/x86/18.log b/measurements/memory/x86/18.log
new file mode 100644
index 0000000..a9879dd
--- /dev/null
+++ b/measurements/memory/x86/18.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   73.09 per second)
+
+16384.00 MiB transferred (74844.08 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2172s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   46.80
+         avg:                                   93.61
+         max:                                  170.30
+         95th percentile:                      150.29
+         sum:                                 1497.83
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1872/0.02
+
diff --git a/measurements/memory/x86/19.log b/measurements/memory/x86/19.log
new file mode 100644
index 0000000..7be3db9
--- /dev/null
+++ b/measurements/memory/x86/19.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   99.55 per second)
+
+16384.00 MiB transferred (101941.22 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1591s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   76.15
+         avg:                                   77.89
+         max:                                   82.07
+         95th percentile:                       81.48
+         sum:                                 1246.20
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1558/0.00
+
diff --git a/measurements/memory/x86/2.log b/measurements/memory/x86/2.log
index e582bbd..c480bd3 100644
--- a/measurements/memory/x86/2.log
+++ b/measurements/memory/x86/2.log
@@ -1,7 +1,7 @@
 sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
 
 Running the test with following options:
-Number of threads: 1
+Number of threads: 8
 Initializing random number generator from current time
 
 
@@ -15,23 +15,23 @@ Initializing worker threads...
 
 Threads started!
 
-Total operations: 20 (   37.63 per second)
+Total operations: 16 (   80.42 per second)
 
-20480.00 MiB transferred (38536.77 MiB/sec)
+16384.00 MiB transferred (82353.79 MiB/sec)
 
 
 General statistics:
-    total time:                          0.5299s
-    total number of events:              20
+    total time:                          0.1972s
+    total number of events:              16
 
 Latency (ms):
-         min:                                   25.63
-         avg:                                   26.48
-         max:                                   36.38
-         95th percentile:                       27.66
-         sum:                                  529.61
+         min:                                   71.20
+         avg:                                   90.45
+         max:                                  117.65
+         95th percentile:                      116.80
+         sum:                                 1447.14
 
 Threads fairness:
-    events (avg/stddev):           20.0000/0.00
-    execution time (avg/stddev):   0.5296/0.00
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1809/0.01
 
diff --git a/measurements/memory/x86/20.log b/measurements/memory/x86/20.log
new file mode 100644
index 0000000..847ed20
--- /dev/null
+++ b/measurements/memory/x86/20.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   93.90 per second)
+
+16384.00 MiB transferred (96155.10 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1687s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   71.25
+         avg:                                   81.27
+         max:                                   90.18
+         95th percentile:                       89.16
+         sum:                                 1300.34
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1625/0.00
+
diff --git a/measurements/memory/x86/21.log b/measurements/memory/x86/21.log
new file mode 100644
index 0000000..6005d9a
--- /dev/null
+++ b/measurements/memory/x86/21.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  113.14 per second)
+
+16384.00 MiB transferred (115856.56 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1398s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   66.87
+         avg:                                   69.44
+         max:                                   72.73
+         95th percentile:                       73.13
+         sum:                                 1111.05
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1389/0.00
+
diff --git a/measurements/memory/x86/22.log b/measurements/memory/x86/22.log
new file mode 100644
index 0000000..f702ede
--- /dev/null
+++ b/measurements/memory/x86/22.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   74.21 per second)
+
+16384.00 MiB transferred (75991.45 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2140s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   55.14
+         avg:                                   91.61
+         max:                                  158.54
+         95th percentile:                      153.02
+         sum:                                 1465.83
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1832/0.02
+
diff --git a/measurements/memory/x86/23.log b/measurements/memory/x86/23.log
new file mode 100644
index 0000000..a1f196c
--- /dev/null
+++ b/measurements/memory/x86/23.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  106.90 per second)
+
+16384.00 MiB transferred (109464.64 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1481s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   66.13
+         avg:                                   72.82
+         max:                                   80.85
+         95th percentile:                       80.03
+         sum:                                 1165.16
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1456/0.00
+
diff --git a/measurements/memory/x86/24.log b/measurements/memory/x86/24.log
new file mode 100644
index 0000000..6dbab72
--- /dev/null
+++ b/measurements/memory/x86/24.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   90.46 per second)
+
+16384.00 MiB transferred (92626.28 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1752s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   38.71
+         avg:                                   74.09
+         max:                                  136.24
+         95th percentile:                      130.13
+         sum:                                 1185.49
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1482/0.02
+
diff --git a/measurements/memory/x86/25.log b/measurements/memory/x86/25.log
new file mode 100644
index 0000000..8889576
--- /dev/null
+++ b/measurements/memory/x86/25.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   75.26 per second)
+
+16384.00 MiB transferred (77065.32 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2109s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   75.17
+         avg:                                   98.76
+         max:                                  118.28
+         95th percentile:                      116.80
+         sum:                                 1580.09
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1975/0.01
+
diff --git a/measurements/memory/x86/26.log b/measurements/memory/x86/26.log
new file mode 100644
index 0000000..c62887e
--- /dev/null
+++ b/measurements/memory/x86/26.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   85.84 per second)
+
+16384.00 MiB transferred (87904.66 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1846s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   46.68
+         avg:                                   77.73
+         max:                                  137.34
+         95th percentile:                      134.90
+         sum:                                 1243.61
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1555/0.02
+
diff --git a/measurements/memory/x86/27.log b/measurements/memory/x86/27.log
new file mode 100644
index 0000000..d34cdf7
--- /dev/null
+++ b/measurements/memory/x86/27.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   68.61 per second)
+
+16384.00 MiB transferred (70260.31 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2315s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   67.82
+         avg:                                  106.58
+         max:                                  163.55
+         95th percentile:                      164.45
+         sum:                                 1705.30
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2132/0.01
+
diff --git a/measurements/memory/x86/28.log b/measurements/memory/x86/28.log
new file mode 100644
index 0000000..0a06093
--- /dev/null
+++ b/measurements/memory/x86/28.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   87.48 per second)
+
+16384.00 MiB transferred (89575.43 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1813s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   69.25
+         avg:                                   89.18
+         max:                                  109.65
+         95th percentile:                      108.68
+         sum:                                 1426.91
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1784/0.00
+
diff --git a/measurements/memory/x86/29.log b/measurements/memory/x86/29.log
new file mode 100644
index 0000000..752f6e6
--- /dev/null
+++ b/measurements/memory/x86/29.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   81.07 per second)
+
+16384.00 MiB transferred (83012.89 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1957s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   77.98
+         avg:                                   93.04
+         max:                                  105.42
+         95th percentile:                      102.97
+         sum:                                 1488.64
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1861/0.01
+
diff --git a/measurements/memory/x86/3.log b/measurements/memory/x86/3.log
index 90efd3b..8538d68 100644
--- a/measurements/memory/x86/3.log
+++ b/measurements/memory/x86/3.log
@@ -1,7 +1,7 @@
 sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
 
 Running the test with following options:
-Number of threads: 1
+Number of threads: 8
 Initializing random number generator from current time
 
 
@@ -15,23 +15,23 @@ Initializing worker threads...
 
 Threads started!
 
-Total operations: 20 (   37.35 per second)
+Total operations: 16 (   97.03 per second)
 
-20480.00 MiB transferred (38249.60 MiB/sec)
+16384.00 MiB transferred (99354.66 MiB/sec)
 
 
 General statistics:
-    total time:                          0.5339s
-    total number of events:              20
+    total time:                          0.1632s
+    total number of events:              16
 
 Latency (ms):
-         min:                                   26.01
-         avg:                                   26.68
-         max:                                   33.21
-         95th percentile:                       27.66
-         sum:                                  533.54
+         min:                                   69.58
+         avg:                                   78.02
+         max:                                   85.93
+         95th percentile:                       86.00
+         sum:                                 1248.33
 
 Threads fairness:
-    events (avg/stddev):           20.0000/0.00
-    execution time (avg/stddev):   0.5335/0.00
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1560/0.00
 
diff --git a/measurements/memory/x86/30.log b/measurements/memory/x86/30.log
new file mode 100644
index 0000000..b2216b5
--- /dev/null
+++ b/measurements/memory/x86/30.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   95.17 per second)
+
+16384.00 MiB transferred (97451.39 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1665s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   69.09
+         avg:                                   79.04
+         max:                                   94.05
+         95th percentile:                       94.10
+         sum:                                 1264.67
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1581/0.01
+
diff --git a/measurements/memory/x86/31.log b/measurements/memory/x86/31.log
new file mode 100644
index 0000000..0b0687c
--- /dev/null
+++ b/measurements/memory/x86/31.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   72.12 per second)
+
+16384.00 MiB transferred (73852.41 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2202s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   81.81
+         avg:                                  107.56
+         max:                                  132.03
+         95th percentile:                      130.13
+         sum:                                 1720.95
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2151/0.00
+
diff --git a/measurements/memory/x86/32.log b/measurements/memory/x86/32.log
new file mode 100644
index 0000000..1725676
--- /dev/null
+++ b/measurements/memory/x86/32.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  100.76 per second)
+
+16384.00 MiB transferred (103176.04 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1571s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   55.36
+         avg:                                   71.95
+         max:                                  101.65
+         95th percentile:                      101.13
+         sum:                                 1151.16
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1439/0.01
+
diff --git a/measurements/memory/x86/33.log b/measurements/memory/x86/33.log
new file mode 100644
index 0000000..306f47b
--- /dev/null
+++ b/measurements/memory/x86/33.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   74.95 per second)
+
+16384.00 MiB transferred (76752.90 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2117s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   89.85
+         avg:                                  103.90
+         max:                                  116.44
+         95th percentile:                      114.72
+         sum:                                 1662.43
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2078/0.00
+
diff --git a/measurements/memory/x86/34.log b/measurements/memory/x86/34.log
new file mode 100644
index 0000000..7aa880b
--- /dev/null
+++ b/measurements/memory/x86/34.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  102.43 per second)
+
+16384.00 MiB transferred (104883.79 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1546s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   73.74
+         avg:                                   75.39
+         max:                                   80.49
+         95th percentile:                       77.19
+         sum:                                 1206.24
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1508/0.00
+
diff --git a/measurements/memory/x86/35.log b/measurements/memory/x86/35.log
new file mode 100644
index 0000000..da62781
--- /dev/null
+++ b/measurements/memory/x86/35.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   79.44 per second)
+
+16384.00 MiB transferred (81349.95 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1996s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   53.35
+         avg:                                   81.47
+         max:                                  146.20
+         95th percentile:                      132.49
+         sum:                                 1303.50
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1629/0.02
+
diff --git a/measurements/memory/x86/36.log b/measurements/memory/x86/36.log
new file mode 100644
index 0000000..ec6a67f
--- /dev/null
+++ b/measurements/memory/x86/36.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  109.14 per second)
+
+16384.00 MiB transferred (111754.38 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1450s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   64.93
+         avg:                                   68.84
+         max:                                   79.89
+         95th percentile:                       70.55
+         sum:                                 1101.39
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1377/0.00
+
diff --git a/measurements/memory/x86/37.log b/measurements/memory/x86/37.log
new file mode 100644
index 0000000..683dcc3
--- /dev/null
+++ b/measurements/memory/x86/37.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   84.17 per second)
+
+16384.00 MiB transferred (86187.34 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1878s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   75.26
+         avg:                                   92.03
+         max:                                  109.92
+         95th percentile:                      108.68
+         sum:                                 1472.49
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1841/0.00
+
diff --git a/measurements/memory/x86/38.log b/measurements/memory/x86/38.log
new file mode 100644
index 0000000..0555595
--- /dev/null
+++ b/measurements/memory/x86/38.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   90.86 per second)
+
+16384.00 MiB transferred (93044.66 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1744s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   67.71
+         avg:                                   83.70
+         max:                                   99.33
+         95th percentile:                       97.55
+         sum:                                 1339.14
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1674/0.00
+
diff --git a/measurements/memory/x86/39.log b/measurements/memory/x86/39.log
new file mode 100644
index 0000000..8e7e4c1
--- /dev/null
+++ b/measurements/memory/x86/39.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   81.17 per second)
+
+16384.00 MiB transferred (83115.57 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1951s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   86.50
+         avg:                                   94.28
+         max:                                  102.49
+         95th percentile:                      101.13
+         sum:                                 1508.50
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1886/0.00
+
diff --git a/measurements/memory/x86/4.log b/measurements/memory/x86/4.log
new file mode 100644
index 0000000..8af7906
--- /dev/null
+++ b/measurements/memory/x86/4.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   92.61 per second)
+
+16384.00 MiB transferred (94828.99 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1712s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   81.35
+         avg:                                   83.98
+         max:                                   88.16
+         95th percentile:                       87.56
+         sum:                                 1343.62
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1680/0.00
+
diff --git a/measurements/memory/x86/40.log b/measurements/memory/x86/40.log
new file mode 100644
index 0000000..62dc94a
--- /dev/null
+++ b/measurements/memory/x86/40.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   72.71 per second)
+
+16384.00 MiB transferred (74459.75 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2184s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   50.76
+         avg:                                   91.52
+         max:                                  167.45
+         95th percentile:                      106.75
+         sum:                                 1464.29
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1830/0.02
+
diff --git a/measurements/memory/x86/41.log b/measurements/memory/x86/41.log
new file mode 100644
index 0000000..11893a8
--- /dev/null
+++ b/measurements/memory/x86/41.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   60.86 per second)
+
+16384.00 MiB transferred (62318.33 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2613s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   78.09
+         avg:                                  119.23
+         max:                                  183.07
+         95th percentile:                      173.58
+         sum:                                 1907.73
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2385/0.02
+
diff --git a/measurements/memory/x86/42.log b/measurements/memory/x86/42.log
new file mode 100644
index 0000000..8d88e57
--- /dev/null
+++ b/measurements/memory/x86/42.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   56.03 per second)
+
+16384.00 MiB transferred (57372.39 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2837s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   61.31
+         avg:                                  125.06
+         max:                                  215.37
+         95th percentile:                      211.60
+         sum:                                 2001.02
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2501/0.02
+
diff --git a/measurements/memory/x86/43.log b/measurements/memory/x86/43.log
new file mode 100644
index 0000000..1b510fe
--- /dev/null
+++ b/measurements/memory/x86/43.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   58.62 per second)
+
+16384.00 MiB transferred (60026.12 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2713s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   63.69
+         avg:                                  119.17
+         max:                                  207.46
+         95th percentile:                      183.21
+         sum:                                 1906.77
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2383/0.02
+
diff --git a/measurements/memory/x86/44.log b/measurements/memory/x86/44.log
new file mode 100644
index 0000000..4314166
--- /dev/null
+++ b/measurements/memory/x86/44.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  102.95 per second)
+
+16384.00 MiB transferred (105422.76 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1538s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   69.90
+         avg:                                   73.58
+         max:                                   83.68
+         95th percentile:                       77.19
+         sum:                                 1177.29
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1472/0.00
+
diff --git a/measurements/memory/x86/45.log b/measurements/memory/x86/45.log
new file mode 100644
index 0000000..75293fa
--- /dev/null
+++ b/measurements/memory/x86/45.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   79.75 per second)
+
+16384.00 MiB transferred (81660.42 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1990s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   71.18
+         avg:                                   91.03
+         max:                                  120.29
+         95th percentile:                      116.80
+         sum:                                 1456.43
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1821/0.01
+
diff --git a/measurements/memory/x86/46.log b/measurements/memory/x86/46.log
new file mode 100644
index 0000000..b56382c
--- /dev/null
+++ b/measurements/memory/x86/46.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   77.34 per second)
+
+16384.00 MiB transferred (79191.68 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2052s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   86.09
+         avg:                                   98.47
+         max:                                  118.97
+         95th percentile:                      118.92
+         sum:                                 1575.56
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1969/0.01
+
diff --git a/measurements/memory/x86/47.log b/measurements/memory/x86/47.log
new file mode 100644
index 0000000..abc8e9b
--- /dev/null
+++ b/measurements/memory/x86/47.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   64.33 per second)
+
+16384.00 MiB transferred (65869.83 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2471s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   84.56
+         avg:                                  114.39
+         max:                                  157.96
+         95th percentile:                      132.49
+         sum:                                 1830.24
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2288/0.01
+
diff --git a/measurements/memory/x86/48.log b/measurements/memory/x86/48.log
new file mode 100644
index 0000000..cc967c8
--- /dev/null
+++ b/measurements/memory/x86/48.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   64.70 per second)
+
+16384.00 MiB transferred (66250.58 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2457s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   91.94
+         avg:                                  116.01
+         max:                                  153.54
+         95th percentile:                      147.61
+         sum:                                 1856.11
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2320/0.01
+
diff --git a/measurements/memory/x86/49.log b/measurements/memory/x86/49.log
new file mode 100644
index 0000000..b7422e8
--- /dev/null
+++ b/measurements/memory/x86/49.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   98.24 per second)
+
+16384.00 MiB transferred (100600.51 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1612s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   62.93
+         avg:                                   74.93
+         max:                                   97.80
+         95th percentile:                       97.55
+         sum:                                 1198.94
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1499/0.01
+
diff --git a/measurements/memory/x86/5.log b/measurements/memory/x86/5.log
new file mode 100644
index 0000000..e1ca283
--- /dev/null
+++ b/measurements/memory/x86/5.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   92.90 per second)
+
+16384.00 MiB transferred (95129.19 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1706s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   77.25
+         avg:                                   82.32
+         max:                                   87.69
+         95th percentile:                       87.56
+         sum:                                 1317.17
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1646/0.00
+
diff --git a/measurements/memory/x86/50.log b/measurements/memory/x86/50.log
new file mode 100644
index 0000000..52ec9df
--- /dev/null
+++ b/measurements/memory/x86/50.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   73.53 per second)
+
+16384.00 MiB transferred (75296.15 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2158s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   88.22
+         avg:                                  106.84
+         max:                                  125.58
+         95th percentile:                      123.28
+         sum:                                 1709.51
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2137/0.00
+
diff --git a/measurements/memory/x86/51.log b/measurements/memory/x86/51.log
new file mode 100644
index 0000000..a78d21c
--- /dev/null
+++ b/measurements/memory/x86/51.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   91.14 per second)
+
+16384.00 MiB transferred (93328.72 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1739s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   71.66
+         avg:                                   84.75
+         max:                                   97.86
+         95th percentile:                       97.55
+         sum:                                 1356.00
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1695/0.00
+
diff --git a/measurements/memory/x86/52.log b/measurements/memory/x86/52.log
new file mode 100644
index 0000000..156bf31
--- /dev/null
+++ b/measurements/memory/x86/52.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   97.96 per second)
+
+16384.00 MiB transferred (100309.26 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1617s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   73.00
+         avg:                                   75.99
+         max:                                   87.22
+         95th percentile:                       77.19
+         sum:                                 1215.88
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1520/0.00
+
diff --git a/measurements/memory/x86/53.log b/measurements/memory/x86/53.log
new file mode 100644
index 0000000..5bec690
--- /dev/null
+++ b/measurements/memory/x86/53.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   80.30 per second)
+
+16384.00 MiB transferred (82231.25 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1976s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   72.09
+         avg:                                   85.76
+         max:                                  121.61
+         95th percentile:                      108.68
+         sum:                                 1372.23
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1715/0.02
+
diff --git a/measurements/memory/x86/54.log b/measurements/memory/x86/54.log
new file mode 100644
index 0000000..5392026
--- /dev/null
+++ b/measurements/memory/x86/54.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  100.45 per second)
+
+16384.00 MiB transferred (102860.54 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1576s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   67.37
+         avg:                                   75.38
+         max:                                   90.13
+         95th percentile:                       84.47
+         sum:                                 1206.09
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1508/0.00
+
diff --git a/measurements/memory/x86/55.log b/measurements/memory/x86/55.log
new file mode 100644
index 0000000..b080f0c
--- /dev/null
+++ b/measurements/memory/x86/55.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  100.98 per second)
+
+16384.00 MiB transferred (103399.91 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1567s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   71.44
+         avg:                                   76.15
+         max:                                   80.74
+         95th percentile:                       80.03
+         sum:                                 1218.39
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1523/0.00
+
diff --git a/measurements/memory/x86/56.log b/measurements/memory/x86/56.log
new file mode 100644
index 0000000..6035a9a
--- /dev/null
+++ b/measurements/memory/x86/56.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   59.11 per second)
+
+16384.00 MiB transferred (60528.41 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2691s
+    total number of events:              16
+
+Latency (ms):
+         min:                                  109.87
+         avg:                                  130.23
+         max:                                  151.46
+         95th percentile:                      137.35
+         sum:                                 2083.66
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2605/0.01
+
diff --git a/measurements/memory/x86/57.log b/measurements/memory/x86/57.log
new file mode 100644
index 0000000..d9f9b48
--- /dev/null
+++ b/measurements/memory/x86/57.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   82.52 per second)
+
+16384.00 MiB transferred (84499.80 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1923s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   76.98
+         avg:                                   91.58
+         max:                                  107.30
+         95th percentile:                      104.84
+         sum:                                 1465.21
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1832/0.01
+
diff --git a/measurements/memory/x86/58.log b/measurements/memory/x86/58.log
new file mode 100644
index 0000000..e84083d
--- /dev/null
+++ b/measurements/memory/x86/58.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  107.72 per second)
+
+16384.00 MiB transferred (110301.47 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1467s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   59.77
+         avg:                                   69.22
+         max:                                   86.69
+         95th percentile:                       86.00
+         sum:                                 1107.51
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1384/0.00
+
diff --git a/measurements/memory/x86/59.log b/measurements/memory/x86/59.log
new file mode 100644
index 0000000..2b135b8
--- /dev/null
+++ b/measurements/memory/x86/59.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   79.12 per second)
+
+16384.00 MiB transferred (81016.71 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2006s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   62.97
+         avg:                                   88.24
+         max:                                  137.28
+         95th percentile:                      137.35
+         sum:                                 1411.77
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1765/0.01
+
diff --git a/measurements/memory/x86/6.log b/measurements/memory/x86/6.log
new file mode 100644
index 0000000..87f180d
--- /dev/null
+++ b/measurements/memory/x86/6.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  103.49 per second)
+
+16384.00 MiB transferred (105978.10 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1530s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   60.23
+         avg:                                   72.64
+         max:                                   92.48
+         95th percentile:                       92.42
+         sum:                                 1162.25
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1453/0.00
+
diff --git a/measurements/memory/x86/60.log b/measurements/memory/x86/60.log
new file mode 100644
index 0000000..59623d5
--- /dev/null
+++ b/measurements/memory/x86/60.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   65.67 per second)
+
+16384.00 MiB transferred (67246.47 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2420s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   67.54
+         avg:                                  104.29
+         max:                                  174.02
+         95th percentile:                      161.51
+         sum:                                 1668.57
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2086/0.03
+
diff --git a/measurements/memory/x86/61.log b/measurements/memory/x86/61.log
new file mode 100644
index 0000000..1d4bdb3
--- /dev/null
+++ b/measurements/memory/x86/61.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   68.65 per second)
+
+16384.00 MiB transferred (70295.82 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2315s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   50.78
+         avg:                                  101.12
+         max:                                  180.36
+         95th percentile:                      167.44
+         sum:                                 1617.89
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2022/0.02
+
diff --git a/measurements/memory/x86/62.log b/measurements/memory/x86/62.log
new file mode 100644
index 0000000..d46ac65
--- /dev/null
+++ b/measurements/memory/x86/62.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  104.42 per second)
+
+16384.00 MiB transferred (106924.11 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1516s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   72.20
+         avg:                                   73.92
+         max:                                   79.13
+         95th percentile:                       75.82
+         sum:                                 1182.73
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1478/0.00
+
diff --git a/measurements/memory/x86/63.log b/measurements/memory/x86/63.log
new file mode 100644
index 0000000..41d9828
--- /dev/null
+++ b/measurements/memory/x86/63.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   77.75 per second)
+
+16384.00 MiB transferred (79616.87 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2042s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   75.27
+         avg:                                   94.69
+         max:                                  112.35
+         95th percentile:                      110.66
+         sum:                                 1515.01
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1894/0.01
+
diff --git a/measurements/memory/x86/64.log b/measurements/memory/x86/64.log
new file mode 100644
index 0000000..57f1905
--- /dev/null
+++ b/measurements/memory/x86/64.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   97.19 per second)
+
+16384.00 MiB transferred (99527.47 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1630s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   76.50
+         avg:                                   78.76
+         max:                                   86.21
+         95th percentile:                       80.03
+         sum:                                 1260.24
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1575/0.00
+
diff --git a/measurements/memory/x86/65.log b/measurements/memory/x86/65.log
new file mode 100644
index 0000000..9398c23
--- /dev/null
+++ b/measurements/memory/x86/65.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  103.22 per second)
+
+16384.00 MiB transferred (105698.74 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1533s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   72.93
+         avg:                                   75.41
+         max:                                   80.29
+         95th percentile:                       77.19
+         sum:                                 1206.61
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1508/0.00
+
diff --git a/measurements/memory/x86/66.log b/measurements/memory/x86/66.log
new file mode 100644
index 0000000..191f5a7
--- /dev/null
+++ b/measurements/memory/x86/66.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  107.62 per second)
+
+16384.00 MiB transferred (110198.57 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1470s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   67.98
+         avg:                                   71.21
+         max:                                   78.78
+         95th percentile:                       78.60
+         sum:                                 1139.32
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1424/0.00
+
diff --git a/measurements/memory/x86/67.log b/measurements/memory/x86/67.log
new file mode 100644
index 0000000..812a343
--- /dev/null
+++ b/measurements/memory/x86/67.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  101.07 per second)
+
+16384.00 MiB transferred (103493.97 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1558s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   75.14
+         avg:                                   76.93
+         max:                                   79.56
+         95th percentile:                       78.60
+         sum:                                 1230.83
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1539/0.00
+
diff --git a/measurements/memory/x86/68.log b/measurements/memory/x86/68.log
new file mode 100644
index 0000000..b20da08
--- /dev/null
+++ b/measurements/memory/x86/68.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   76.38 per second)
+
+16384.00 MiB transferred (78212.49 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2078s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   71.66
+         avg:                                  100.03
+         max:                                  126.82
+         95th percentile:                      125.52
+         sum:                                 1600.41
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2001/0.00
+
diff --git a/measurements/memory/x86/69.log b/measurements/memory/x86/69.log
new file mode 100644
index 0000000..08755aa
--- /dev/null
+++ b/measurements/memory/x86/69.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   93.84 per second)
+
+16384.00 MiB transferred (96089.15 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1689s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   56.05
+         avg:                                   75.88
+         max:                                  112.65
+         95th percentile:                      110.66
+         sum:                                 1214.13
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1518/0.01
+
diff --git a/measurements/memory/x86/7.log b/measurements/memory/x86/7.log
new file mode 100644
index 0000000..71a9eda
--- /dev/null
+++ b/measurements/memory/x86/7.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   80.69 per second)
+
+16384.00 MiB transferred (82629.39 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1967s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   60.99
+         avg:                                   83.05
+         max:                                  135.57
+         95th percentile:                      130.13
+         sum:                                 1328.87
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1661/0.02
+
diff --git a/measurements/memory/x86/70.log b/measurements/memory/x86/70.log
new file mode 100644
index 0000000..362ca23
--- /dev/null
+++ b/measurements/memory/x86/70.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   67.80 per second)
+
+16384.00 MiB transferred (69422.44 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2343s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   98.37
+         avg:                                  114.84
+         max:                                  130.48
+         95th percentile:                      130.13
+         sum:                                 1837.43
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2297/0.00
+
diff --git a/measurements/memory/x86/71.log b/measurements/memory/x86/71.log
new file mode 100644
index 0000000..c611256
--- /dev/null
+++ b/measurements/memory/x86/71.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   75.79 per second)
+
+16384.00 MiB transferred (77610.77 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2095s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   39.52
+         avg:                                   89.44
+         max:                                  169.73
+         95th percentile:                      142.39
+         sum:                                 1431.06
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1789/0.02
+
diff --git a/measurements/memory/x86/72.log b/measurements/memory/x86/72.log
new file mode 100644
index 0000000..597887a
--- /dev/null
+++ b/measurements/memory/x86/72.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   92.77 per second)
+
+16384.00 MiB transferred (94996.97 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1707s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   70.15
+         avg:                                   79.20
+         max:                                   95.24
+         95th percentile:                       95.81
+         sum:                                 1267.15
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1584/0.01
+
diff --git a/measurements/memory/x86/73.log b/measurements/memory/x86/73.log
new file mode 100644
index 0000000..cf96af7
--- /dev/null
+++ b/measurements/memory/x86/73.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  116.60 per second)
+
+16384.00 MiB transferred (119396.61 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1355s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   65.06
+         avg:                                   66.95
+         max:                                   70.12
+         95th percentile:                       70.55
+         sum:                                 1071.20
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1339/0.00
+
diff --git a/measurements/memory/x86/74.log b/measurements/memory/x86/74.log
new file mode 100644
index 0000000..0aebcff
--- /dev/null
+++ b/measurements/memory/x86/74.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   80.07 per second)
+
+16384.00 MiB transferred (81996.65 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1982s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   80.41
+         avg:                                   94.63
+         max:                                  107.04
+         95th percentile:                      106.75
+         sum:                                 1514.10
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1893/0.00
+
diff --git a/measurements/memory/x86/75.log b/measurements/memory/x86/75.log
new file mode 100644
index 0000000..bee58ff
--- /dev/null
+++ b/measurements/memory/x86/75.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   63.34 per second)
+
+16384.00 MiB transferred (64858.00 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2510s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   61.57
+         avg:                                  111.55
+         max:                                  189.27
+         95th percentile:                      144.97
+         sum:                                 1784.77
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2231/0.02
+
diff --git a/measurements/memory/x86/76.log b/measurements/memory/x86/76.log
new file mode 100644
index 0000000..bca2b65
--- /dev/null
+++ b/measurements/memory/x86/76.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  103.91 per second)
+
+16384.00 MiB transferred (106405.03 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1523s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   73.36
+         avg:                                   74.61
+         max:                                   77.00
+         95th percentile:                       75.82
+         sum:                                 1193.76
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1492/0.00
+
diff --git a/measurements/memory/x86/77.log b/measurements/memory/x86/77.log
new file mode 100644
index 0000000..536162e
--- /dev/null
+++ b/measurements/memory/x86/77.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  100.33 per second)
+
+16384.00 MiB transferred (102732.83 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1578s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   73.33
+         avg:                                   75.81
+         max:                                   83.97
+         95th percentile:                       80.03
+         sum:                                 1212.89
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1516/0.00
+
diff --git a/measurements/memory/x86/78.log b/measurements/memory/x86/78.log
new file mode 100644
index 0000000..0736528
--- /dev/null
+++ b/measurements/memory/x86/78.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   91.10 per second)
+
+16384.00 MiB transferred (93284.55 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1739s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   67.50
+         avg:                                   82.69
+         max:                                   97.38
+         95th percentile:                       95.81
+         sum:                                 1323.05
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1654/0.00
+
diff --git a/measurements/memory/x86/79.log b/measurements/memory/x86/79.log
new file mode 100644
index 0000000..167349f
--- /dev/null
+++ b/measurements/memory/x86/79.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   58.18 per second)
+
+16384.00 MiB transferred (59578.81 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2733s
+    total number of events:              16
+
+Latency (ms):
+         min:                                  113.61
+         avg:                                  132.06
+         max:                                  159.52
+         95th percentile:                      158.63
+         sum:                                 2112.92
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2641/0.01
+
diff --git a/measurements/memory/x86/8.log b/measurements/memory/x86/8.log
new file mode 100644
index 0000000..22df35c
--- /dev/null
+++ b/measurements/memory/x86/8.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   77.95 per second)
+
+16384.00 MiB transferred (79822.28 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2036s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   63.15
+         avg:                                   95.63
+         max:                                  119.12
+         95th percentile:                      114.72
+         sum:                                 1530.04
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1913/0.01
+
diff --git a/measurements/memory/x86/80.log b/measurements/memory/x86/80.log
new file mode 100644
index 0000000..fbceb76
--- /dev/null
+++ b/measurements/memory/x86/80.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   68.49 per second)
+
+16384.00 MiB transferred (70135.44 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2320s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   95.17
+         avg:                                  112.71
+         max:                                  127.46
+         95th percentile:                      127.81
+         sum:                                 1803.40
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2254/0.00
+
diff --git a/measurements/memory/x86/81.log b/measurements/memory/x86/81.log
new file mode 100644
index 0000000..bd6b5e6
--- /dev/null
+++ b/measurements/memory/x86/81.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  116.86 per second)
+
+16384.00 MiB transferred (119668.49 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1351s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   64.27
+         avg:                                   66.49
+         max:                                   70.74
+         95th percentile:                       70.55
+         sum:                                 1063.86
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1330/0.00
+
diff --git a/measurements/memory/x86/82.log b/measurements/memory/x86/82.log
new file mode 100644
index 0000000..8d918d6
--- /dev/null
+++ b/measurements/memory/x86/82.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   95.62 per second)
+
+16384.00 MiB transferred (97915.95 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1657s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   70.96
+         avg:                                   78.06
+         max:                                   94.60
+         95th percentile:                       94.10
+         sum:                                 1249.03
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1561/0.01
+
diff --git a/measurements/memory/x86/83.log b/measurements/memory/x86/83.log
new file mode 100644
index 0000000..799b987
--- /dev/null
+++ b/measurements/memory/x86/83.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   69.73 per second)
+
+16384.00 MiB transferred (71398.50 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2278s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   84.25
+         avg:                                  111.48
+         max:                                  136.03
+         95th percentile:                      134.90
+         sum:                                 1783.73
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2230/0.00
+
diff --git a/measurements/memory/x86/84.log b/measurements/memory/x86/84.log
new file mode 100644
index 0000000..9baaff8
--- /dev/null
+++ b/measurements/memory/x86/84.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  101.15 per second)
+
+16384.00 MiB transferred (103577.94 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1562s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   70.07
+         avg:                                   75.75
+         max:                                   80.38
+         95th percentile:                       80.03
+         sum:                                 1211.98
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1515/0.00
+
diff --git a/measurements/memory/x86/85.log b/measurements/memory/x86/85.log
new file mode 100644
index 0000000..f40ee83
--- /dev/null
+++ b/measurements/memory/x86/85.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   65.93 per second)
+
+16384.00 MiB transferred (67511.91 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2411s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   73.98
+         avg:                                  109.55
+         max:                                  159.84
+         95th percentile:                      139.85
+         sum:                                 1752.80
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2191/0.01
+
diff --git a/measurements/memory/x86/86.log b/measurements/memory/x86/86.log
new file mode 100644
index 0000000..e128580
--- /dev/null
+++ b/measurements/memory/x86/86.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   81.84 per second)
+
+16384.00 MiB transferred (83802.42 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1939s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   55.68
+         avg:                                   86.71
+         max:                                  137.93
+         95th percentile:                      108.68
+         sum:                                 1387.42
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1734/0.01
+
diff --git a/measurements/memory/x86/87.log b/measurements/memory/x86/87.log
new file mode 100644
index 0000000..13ceda5
--- /dev/null
+++ b/measurements/memory/x86/87.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   98.85 per second)
+
+16384.00 MiB transferred (101222.95 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1602s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   72.19
+         avg:                                   77.10
+         max:                                   87.91
+         95th percentile:                       84.47
+         sum:                                 1233.67
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1542/0.00
+
diff --git a/measurements/memory/x86/88.log b/measurements/memory/x86/88.log
new file mode 100644
index 0000000..fe73044
--- /dev/null
+++ b/measurements/memory/x86/88.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   85.71 per second)
+
+16384.00 MiB transferred (87768.83 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1849s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   75.80
+         avg:                                   85.54
+         max:                                  106.01
+         95th percentile:                       99.33
+         sum:                                 1368.70
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1711/0.01
+
diff --git a/measurements/memory/x86/89.log b/measurements/memory/x86/89.log
new file mode 100644
index 0000000..f1f34b6
--- /dev/null
+++ b/measurements/memory/x86/89.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   98.25 per second)
+
+16384.00 MiB transferred (100612.47 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1610s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   71.79
+         avg:                                   76.11
+         max:                                   88.03
+         95th percentile:                       87.56
+         sum:                                 1217.82
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1522/0.01
+
diff --git a/measurements/memory/x86/9.log b/measurements/memory/x86/9.log
new file mode 100644
index 0000000..e2858b5
--- /dev/null
+++ b/measurements/memory/x86/9.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   75.63 per second)
+
+16384.00 MiB transferred (77441.55 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2099s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   65.96
+         avg:                                   88.37
+         max:                                  143.80
+         95th percentile:                      139.85
+         sum:                                 1413.88
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1767/0.03
+
diff --git a/measurements/memory/x86/90.log b/measurements/memory/x86/90.log
new file mode 100644
index 0000000..266c126
--- /dev/null
+++ b/measurements/memory/x86/90.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   81.70 per second)
+
+16384.00 MiB transferred (83664.49 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1940s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   59.41
+         avg:                                   78.25
+         max:                                  134.42
+         95th percentile:                      130.13
+         sum:                                 1251.98
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1565/0.02
+
diff --git a/measurements/memory/x86/91.log b/measurements/memory/x86/91.log
new file mode 100644
index 0000000..83dd01b
--- /dev/null
+++ b/measurements/memory/x86/91.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   99.77 per second)
+
+16384.00 MiB transferred (102164.95 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1588s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   72.55
+         avg:                                   76.45
+         max:                                   85.32
+         95th percentile:                       80.03
+         sum:                                 1223.26
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1529/0.00
+
diff --git a/measurements/memory/x86/92.log b/measurements/memory/x86/92.log
new file mode 100644
index 0000000..48eab6d
--- /dev/null
+++ b/measurements/memory/x86/92.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   82.26 per second)
+
+16384.00 MiB transferred (84232.58 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1929s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   82.21
+         avg:                                   94.02
+         max:                                  102.11
+         95th percentile:                      101.13
+         sum:                                 1504.29
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1880/0.00
+
diff --git a/measurements/memory/x86/93.log b/measurements/memory/x86/93.log
new file mode 100644
index 0000000..24980a2
--- /dev/null
+++ b/measurements/memory/x86/93.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   91.24 per second)
+
+16384.00 MiB transferred (93426.00 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1737s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   47.90
+         avg:                                   73.47
+         max:                                  125.72
+         95th percentile:                      121.08
+         sum:                                 1175.48
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1469/0.02
+
diff --git a/measurements/memory/x86/94.log b/measurements/memory/x86/94.log
new file mode 100644
index 0000000..e2f0fed
--- /dev/null
+++ b/measurements/memory/x86/94.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   69.53 per second)
+
+16384.00 MiB transferred (71199.69 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2285s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   63.70
+         avg:                                   99.28
+         max:                                  164.51
+         95th percentile:                      155.80
+         sum:                                 1588.48
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1986/0.02
+
diff --git a/measurements/memory/x86/95.log b/measurements/memory/x86/95.log
new file mode 100644
index 0000000..7fd015e
--- /dev/null
+++ b/measurements/memory/x86/95.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (  104.42 per second)
+
+16384.00 MiB transferred (106925.10 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1516s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   65.47
+         avg:                                   70.92
+         max:                                   86.03
+         95th percentile:                       71.83
+         sum:                                 1134.76
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1418/0.00
+
diff --git a/measurements/memory/x86/96.log b/measurements/memory/x86/96.log
new file mode 100644
index 0000000..892d81b
--- /dev/null
+++ b/measurements/memory/x86/96.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   71.77 per second)
+
+16384.00 MiB transferred (73490.57 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2213s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   79.58
+         avg:                                  105.45
+         max:                                  125.14
+         95th percentile:                      125.52
+         sum:                                 1687.27
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2109/0.01
+
diff --git a/measurements/memory/x86/97.log b/measurements/memory/x86/97.log
new file mode 100644
index 0000000..f1163d6
--- /dev/null
+++ b/measurements/memory/x86/97.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   90.32 per second)
+
+16384.00 MiB transferred (92483.77 MiB/sec)
+
+
+General statistics:
+    total time:                          0.1755s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   69.71
+         avg:                                   82.12
+         max:                                   97.22
+         95th percentile:                       95.81
+         sum:                                 1313.87
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1642/0.01
+
diff --git a/measurements/memory/x86/98.log b/measurements/memory/x86/98.log
new file mode 100644
index 0000000..f7cb680
--- /dev/null
+++ b/measurements/memory/x86/98.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   74.54 per second)
+
+16384.00 MiB transferred (76330.40 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2130s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   49.36
+         avg:                                   90.31
+         max:                                  163.46
+         95th percentile:                      123.28
+         sum:                                 1444.96
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.1806/0.02
+
diff --git a/measurements/memory/x86/99.log b/measurements/memory/x86/99.log
new file mode 100644
index 0000000..a7ce2c5
--- /dev/null
+++ b/measurements/memory/x86/99.log
@@ -0,0 +1,37 @@
+sysbench 1.0.20 (using system LuaJIT 2.1.1761786044)
+
+Running the test with following options:
+Number of threads: 8
+Initializing random number generator from current time
+
+
+Running memory speed test with the following options:
+  block size: 1048576KiB
+  total size: 20480MiB
+  operation: read
+  scope: global
+
+Initializing worker threads...
+
+Threads started!
+
+Total operations: 16 (   68.00 per second)
+
+16384.00 MiB transferred (69636.56 MiB/sec)
+
+
+General statistics:
+    total time:                          0.2337s
+    total number of events:              16
+
+Latency (ms):
+         min:                                   87.07
+         avg:                                  111.85
+         max:                                  131.42
+         95th percentile:                      130.13
+         sum:                                 1789.66
+
+Threads fairness:
+    events (avg/stddev):           2.0000/0.00
+    execution time (avg/stddev):   0.2237/0.01
+
diff --git a/measurements/variant_baseline/x86/3.log b/measurements/variant_baseline/x86/3.log
new file mode 100644
index 0000000..f3a1d6f
--- /dev/null
+++ b/measurements/variant_baseline/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_baseline.cpp -o main_baseline -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 27602905.00 cycles
+P(n): 1.33 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 108346220.00 cycles
+P(n): 1.34 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 436171900.00 cycles
+P(n): 1.33 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 2815330460.00 cycles
+P(n): 1.28 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 11366685120.00 cycles
+P(n): 1.27 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 46485487760.00 cycles
+P(n): 1.24 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 312273688920.00 cycles
+P(n): 1.15 flops/cycle
+
diff --git a/measurements/variant_branchless/x86/2.log b/measurements/variant_branchless/x86/2.log
new file mode 100644
index 0000000..8da19dc
--- /dev/null
+++ b/measurements/variant_branchless/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_branchless.cpp -o main_branchless -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 32618855.00 cycles
+P(n): 1.13 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 129903020.00 cycles
+P(n): 1.12 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 546607480.00 cycles
+P(n): 1.06 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 3564103380.00 cycles
+P(n): 1.01 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 14460744480.00 cycles
+P(n): 1.00 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 59317253600.00 cycles
+P(n): 0.97 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 386083944780.00 cycles
+P(n): 0.93 flops/cycle
+
diff --git a/measurements/variant_branchless/x86/3.log b/measurements/variant_branchless/x86/3.log
new file mode 100644
index 0000000..6d6f654
--- /dev/null
+++ b/measurements/variant_branchless/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_branchless.cpp -o main_branchless -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 32549760.00 cycles
+P(n): 1.13 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 132472520.00 cycles
+P(n): 1.10 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 546899100.00 cycles
+P(n): 1.06 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 3595898300.00 cycles
+P(n): 1.00 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 14398657780.00 cycles
+P(n): 1.00 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 59074417020.00 cycles
+P(n): 0.98 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 384714694000.00 cycles
+P(n): 0.94 flops/cycle
+
diff --git a/measurements/variant_branchless_sqrtfree/x86/2.log b/measurements/variant_branchless_sqrtfree/x86/2.log
new file mode 100644
index 0000000..7072d5b
--- /dev/null
+++ b/measurements/variant_branchless_sqrtfree/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_branchless_sqrtfree.cpp -o main_branchless_sqrtfree -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 32048610.00 cycles
+P(n): 1.15 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 124926140.00 cycles
+P(n): 1.16 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 496226700.00 cycles
+P(n): 1.17 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 3098696200.00 cycles
+P(n): 1.16 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 12321978700.00 cycles
+P(n): 1.17 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 49235152160.00 cycles
+P(n): 1.17 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 307537206280.00 cycles
+P(n): 1.17 flops/cycle
+
diff --git a/measurements/variant_branchless_sqrtfree/x86/3.log b/measurements/variant_branchless_sqrtfree/x86/3.log
new file mode 100644
index 0000000..91b16f6
--- /dev/null
+++ b/measurements/variant_branchless_sqrtfree/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_branchless_sqrtfree.cpp -o main_branchless_sqrtfree -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 31528725.00 cycles
+P(n): 1.17 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 124375380.00 cycles
+P(n): 1.17 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 494656240.00 cycles
+P(n): 1.17 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 3069950280.00 cycles
+P(n): 1.18 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 12264064020.00 cycles
+P(n): 1.18 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 49136256080.00 cycles
+P(n): 1.17 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 310316526840.00 cycles
+P(n): 1.16 flops/cycle
+
diff --git a/measurements/variant_cutoff/x86/2.log b/measurements/variant_cutoff/x86/2.log
new file mode 100644
index 0000000..e4aef82
--- /dev/null
+++ b/measurements/variant_cutoff/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_cutoff.cpp -o main_cutoff -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 17857572.50 cycles
+P(n): 2.06 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 76450530.00 cycles
+P(n): 1.90 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 270167300.00 cycles
+P(n): 2.14 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 1898509880.00 cycles
+P(n): 1.90 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 7707918240.00 cycles
+P(n): 1.87 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 31905756600.00 cycles
+P(n): 1.81 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 215725945760.00 cycles
+P(n): 1.67 flops/cycle
+
diff --git a/measurements/variant_cutoff/x86/3.log b/measurements/variant_cutoff/x86/3.log
new file mode 100644
index 0000000..887fa38
--- /dev/null
+++ b/measurements/variant_cutoff/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_cutoff.cpp -o main_cutoff -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 17637622.50 cycles
+P(n): 2.08 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 76060470.00 cycles
+P(n): 1.91 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 267233860.00 cycles
+P(n): 2.17 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 1897077760.00 cycles
+P(n): 1.90 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 7759546840.00 cycles
+P(n): 1.86 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 32218084860.00 cycles
+P(n): 1.79 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 216623652880.00 cycles
+P(n): 1.66 flops/cycle
+
diff --git a/measurements/variant_cutoff_callsite/x86/2.log b/measurements/variant_cutoff_callsite/x86/2.log
new file mode 100644
index 0000000..29a7e24
--- /dev/null
+++ b/measurements/variant_cutoff_callsite/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_cutoff_callsite.cpp -o main_cutoff_callsite -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 16934235.00 cycles
+P(n): 2.17 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 71336000.00 cycles
+P(n): 2.04 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 260955100.00 cycles
+P(n): 2.22 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 1819642180.00 cycles
+P(n): 1.98 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 7536435560.00 cycles
+P(n): 1.91 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 30984442660.00 cycles
+P(n): 1.86 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 210935037300.00 cycles
+P(n): 1.71 flops/cycle
+
diff --git a/measurements/variant_cutoff_callsite/x86/3.log b/measurements/variant_cutoff_callsite/x86/3.log
new file mode 100644
index 0000000..62f62de
--- /dev/null
+++ b/measurements/variant_cutoff_callsite/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_cutoff_callsite.cpp -o main_cutoff_callsite -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 16651247.50 cycles
+P(n): 2.21 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 71732010.00 cycles
+P(n): 2.03 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 251676200.00 cycles
+P(n): 2.30 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 1824751540.00 cycles
+P(n): 1.98 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 7534812080.00 cycles
+P(n): 1.91 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 31129863960.00 cycles
+P(n): 1.85 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 210111990180.00 cycles
+P(n): 1.71 flops/cycle
+
diff --git a/measurements/variant_float_baseline/x86/2.log b/measurements/variant_float_baseline/x86/2.log
new file mode 100644
index 0000000..15682ec
--- /dev/null
+++ b/measurements/variant_float_baseline/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_baseline.cpp -o main_float_baseline -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 23906564.00 cycles
+P(n): 1.54 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 94126864.00 cycles
+P(n): 1.55 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 379665728.00 cycles
+P(n): 1.52 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 2738376960.00 cycles
+P(n): 1.32 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 11763845120.00 cycles
+P(n): 1.23 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 51998625792.00 cycles
+P(n): 1.11 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 369826004992.00 cycles
+P(n): 0.97 flops/cycle
+
diff --git a/measurements/variant_float_baseline/x86/3.log b/measurements/variant_float_baseline/x86/3.log
new file mode 100644
index 0000000..70e84bf
--- /dev/null
+++ b/measurements/variant_float_baseline/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_baseline.cpp -o main_float_baseline -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 23530572.00 cycles
+P(n): 1.56 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 93875008.00 cycles
+P(n): 1.55 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 380102016.00 cycles
+P(n): 1.52 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 2731053824.00 cycles
+P(n): 1.32 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 11666406400.00 cycles
+P(n): 1.24 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 51928870912.00 cycles
+P(n): 1.11 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 370410651648.00 cycles
+P(n): 0.97 flops/cycle
+
diff --git a/measurements/variant_float_inlining/x86/2.log b/measurements/variant_float_inlining/x86/2.log
new file mode 100644
index 0000000..054a921
--- /dev/null
+++ b/measurements/variant_float_inlining/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining.cpp -o main_float_inlining -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 20872928.00 cycles
+P(n): 1.76 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 83152760.00 cycles
+P(n): 1.75 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 334965856.00 cycles
+P(n): 1.73 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 2470004736.00 cycles
+P(n): 1.46 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 10627587072.00 cycles
+P(n): 1.36 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 47728148480.00 cycles
+P(n): 1.21 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 342776479744.00 cycles
+P(n): 1.05 flops/cycle
+
diff --git a/measurements/variant_float_inlining/x86/3.log b/measurements/variant_float_inlining/x86/3.log
new file mode 100644
index 0000000..d041d11
--- /dev/null
+++ b/measurements/variant_float_inlining/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining.cpp -o main_float_inlining -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 21161456.00 cycles
+P(n): 1.74 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 84570152.00 cycles
+P(n): 1.72 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 342739392.00 cycles
+P(n): 1.69 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 2469612800.00 cycles
+P(n): 1.46 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 10730601472.00 cycles
+P(n): 1.34 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 47785230336.00 cycles
+P(n): 1.21 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 342750101504.00 cycles
+P(n): 1.05 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA/x86/2.log b/measurements/variant_float_inlining_SoA/x86/2.log
new file mode 100644
index 0000000..8299833
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA.cpp -o main_float_inlining_SoA -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 20983524.00 cycles
+P(n): 1.75 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 85081848.00 cycles
+P(n): 1.71 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 337773472.00 cycles
+P(n): 1.71 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 2496699136.00 cycles
+P(n): 1.44 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 10822607872.00 cycles
+P(n): 1.33 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 48131854336.00 cycles
+P(n): 1.20 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 346968686592.00 cycles
+P(n): 1.04 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA/x86/3.log b/measurements/variant_float_inlining_SoA/x86/3.log
new file mode 100644
index 0000000..92e5084
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA.cpp -o main_float_inlining_SoA -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 21201458.00 cycles
+P(n): 1.73 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 84348232.00 cycles
+P(n): 1.72 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 338714912.00 cycles
+P(n): 1.71 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 2491379712.00 cycles
+P(n): 1.45 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 10827898880.00 cycles
+P(n): 1.33 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 48295981056.00 cycles
+P(n): 1.19 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 346709753856.00 cycles
+P(n): 1.04 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX2_polyexp_expfix/x86/1.log b/measurements/variant_float_inlining_SoA_AVX2_polyexp_expfix/x86/1.log
new file mode 100644
index 0000000..6062c1a
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX2_polyexp_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX2_polyexp_expfix.cpp -o main_float_inlining_SoA_AVX2_polyexp_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 5860437.00 cycles
+P(n): 6.27 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 19736818.00 cycles
+P(n): 7.37 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 71295200.00 cycles
+P(n): 8.12 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 434200736.00 cycles
+P(n): 8.31 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1690932352.00 cycles
+P(n): 8.52 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 6726832128.00 cycles
+P(n): 8.57 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 41595662336.00 cycles
+P(n): 8.66 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 166348996608.00 cycles
+P(n): 3.49 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX2_polyexp_expfix/x86/2.log b/measurements/variant_float_inlining_SoA_AVX2_polyexp_expfix/x86/2.log
new file mode 100644
index 0000000..652e0a6
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX2_polyexp_expfix/x86/2.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX2_polyexp_expfix.cpp -o main_float_inlining_SoA_AVX2_polyexp_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 5885449.00 cycles
+P(n): 6.24 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 19775262.00 cycles
+P(n): 7.36 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 71759640.00 cycles
+P(n): 8.07 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 442073856.00 cycles
+P(n): 8.16 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1698162944.00 cycles
+P(n): 8.49 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 6719285248.00 cycles
+P(n): 8.58 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 41795964928.00 cycles
+P(n): 8.62 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 166340608000.00 cycles
+P(n): 3.49 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX2_polyexp_expfix/x86/3.log b/measurements/variant_float_inlining_SoA_AVX2_polyexp_expfix/x86/3.log
new file mode 100644
index 0000000..543ba85
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX2_polyexp_expfix/x86/3.log
@@ -0,0 +1,64 @@
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 5839795.00 cycles
+P(n): 6.29 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 19706848.00 cycles
+P(n): 7.38 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 71913632.00 cycles
+P(n): 8.05 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 431199296.00 cycles
+P(n): 8.37 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1710698496.00 cycles
+P(n): 8.43 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 6773665792.00 cycles
+P(n): 8.51 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 43346759680.00 cycles
+P(n): 8.31 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 169504800768.00 cycles
+P(n): 3.43 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix/x86/1.log b/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix/x86/1.log
new file mode 100644
index 0000000..7e95d51
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix.cpp -o main_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 5075402.00 cycles
+P(n): 7.24 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 16874942.00 cycles
+P(n): 8.62 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 60586048.00 cycles
+P(n): 9.56 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 363991840.00 cycles
+P(n): 9.91 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1424352768.00 cycles
+P(n): 10.12 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 5638619648.00 cycles
+P(n): 10.22 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 35081228288.00 cycles
+P(n): 10.26 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 140734889984.00 cycles
+P(n): 4.13 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix/x86/2.log b/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix/x86/2.log
new file mode 100644
index 0000000..322109b
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix/x86/2.log
@@ -0,0 +1,64 @@
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 5095124.00 cycles
+P(n): 7.21 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 16958378.00 cycles
+P(n): 8.58 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 60894740.00 cycles
+P(n): 9.51 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 364922464.00 cycles
+P(n): 9.89 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1432131456.00 cycles
+P(n): 10.07 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 5669585408.00 cycles
+P(n): 10.16 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 35706060800.00 cycles
+P(n): 10.08 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 143150546944.00 cycles
+P(n): 4.06 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix/x86/3.log b/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix/x86/3.log
new file mode 100644
index 0000000..a1c9040
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix/x86/3.log
@@ -0,0 +1,64 @@
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 5107488.00 cycles
+P(n): 7.19 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 16972588.00 cycles
+P(n): 8.57 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 61120552.00 cycles
+P(n): 9.47 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 375650720.00 cycles
+P(n): 9.60 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1444334464.00 cycles
+P(n): 9.98 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 5717088768.00 cycles
+P(n): 10.08 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 36025249792.00 cycles
+P(n): 10.00 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 143378251776.00 cycles
+P(n): 4.05 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix/x86/1.log b/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix/x86/1.log
new file mode 100644
index 0000000..eab143b
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix.cpp -o main_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4870774.50 cycles
+P(n): 7.54 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 15640118.00 cycles
+P(n): 9.30 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 55310808.00 cycles
+P(n): 10.47 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 329563872.00 cycles
+P(n): 10.95 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1283268352.00 cycles
+P(n): 11.23 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 5083250688.00 cycles
+P(n): 11.34 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 31586570240.00 cycles
+P(n): 11.40 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 126882955264.00 cycles
+P(n): 4.58 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix/x86/2.log b/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix/x86/2.log
new file mode 100644
index 0000000..e4dcd7a
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix/x86/2.log
@@ -0,0 +1,64 @@
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4872185.50 cycles
+P(n): 7.54 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 15664138.00 cycles
+P(n): 9.29 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 55368740.00 cycles
+P(n): 10.46 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 329987008.00 cycles
+P(n): 10.93 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1293044736.00 cycles
+P(n): 11.15 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 5130688512.00 cycles
+P(n): 11.23 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 32060348416.00 cycles
+P(n): 11.23 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 129057972224.00 cycles
+P(n): 4.50 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix/x86/3.log b/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix/x86/3.log
new file mode 100644
index 0000000..da0c24a
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix/x86/3.log
@@ -0,0 +1,64 @@
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4901788.00 cycles
+P(n): 7.50 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 15669965.00 cycles
+P(n): 9.28 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 55911480.00 cycles
+P(n): 10.35 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 338225088.00 cycles
+P(n): 10.67 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1302161408.00 cycles
+P(n): 11.07 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 5189515776.00 cycles
+P(n): 11.11 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 32577761280.00 cycles
+P(n): 11.05 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 129035010048.00 cycles
+P(n): 4.50 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_expfix/x86/1.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_expfix/x86/1.log
new file mode 100644
index 0000000..533d3a6
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4618692.00 cycles
+P(n): 7.95 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14020708.00 cycles
+P(n): 10.38 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 47426524.00 cycles
+P(n): 12.21 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 271672832.00 cycles
+P(n): 13.28 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1041892096.00 cycles
+P(n): 13.84 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 4086357760.00 cycles
+P(n): 14.10 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 25345890304.00 cycles
+P(n): 14.21 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 100797374464.00 cycles
+P(n): 5.77 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_expfix/x86/2.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_expfix/x86/2.log
new file mode 100644
index 0000000..1d0ae94
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_expfix/x86/2.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4660804.50 cycles
+P(n): 7.88 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14015782.00 cycles
+P(n): 10.38 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 47630040.00 cycles
+P(n): 12.16 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 272007200.00 cycles
+P(n): 13.26 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1042778816.00 cycles
+P(n): 13.82 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 4092400640.00 cycles
+P(n): 14.08 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 25352386560.00 cycles
+P(n): 14.20 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 101285814272.00 cycles
+P(n): 5.74 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_expfix/x86/3.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_expfix/x86/3.log
new file mode 100644
index 0000000..49a7e56
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_expfix/x86/3.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4633135.50 cycles
+P(n): 7.93 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14098810.00 cycles
+P(n): 10.32 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 47519168.00 cycles
+P(n): 12.18 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 272423904.00 cycles
+P(n): 13.24 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1060791424.00 cycles
+P(n): 13.59 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 4102831872.00 cycles
+P(n): 14.05 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 25357774848.00 cycles
+P(n): 14.20 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 100551254016.00 cycles
+P(n): 5.78 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/1.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/1.log
new file mode 100644
index 0000000..7cc9274
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4899351.00 cycles
+P(n): 7.50 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14704840.00 cycles
+P(n): 9.89 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 48870456.00 cycles
+P(n): 11.85 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 274548320.00 cycles
+P(n): 13.14 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1067638016.00 cycles
+P(n): 13.50 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 4133780736.00 cycles
+P(n): 13.94 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 25536294912.00 cycles
+P(n): 14.10 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 101531860992.00 cycles
+P(n): 5.72 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/2.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/2.log
new file mode 100644
index 0000000..b038931
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/2.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4906050.00 cycles
+P(n): 7.49 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14798538.00 cycles
+P(n): 9.83 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 48817384.00 cycles
+P(n): 11.86 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 276943584.00 cycles
+P(n): 13.03 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1052295168.00 cycles
+P(n): 13.70 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 4133724672.00 cycles
+P(n): 13.94 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 25570220032.00 cycles
+P(n): 14.08 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 101490999296.00 cycles
+P(n): 5.73 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/3.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/3.log
new file mode 100644
index 0000000..ed48881
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/3.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4896082.00 cycles
+P(n): 7.50 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14589518.00 cycles
+P(n): 9.97 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 48579536.00 cycles
+P(n): 11.92 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 275277376.00 cycles
+P(n): 13.10 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1050094400.00 cycles
+P(n): 13.73 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 4110344448.00 cycles
+P(n): 14.02 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 25639516160.00 cycles
+P(n): 14.04 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 101213003776.00 cycles
+P(n): 5.74 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix/x86/1.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix/x86/1.log
new file mode 100644
index 0000000..29aad47
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4444855.00 cycles
+P(n): 8.27 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 13834708.00 cycles
+P(n): 10.52 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 47410956.00 cycles
+P(n): 12.21 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 276381408.00 cycles
+P(n): 13.05 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1067362496.00 cycles
+P(n): 13.51 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 4156325376.00 cycles
+P(n): 13.87 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 26170245120.00 cycles
+P(n): 13.76 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 104163885056.00 cycles
+P(n): 5.58 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix/x86/2.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix/x86/2.log
new file mode 100644
index 0000000..dcd470c
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix/x86/2.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4415419.00 cycles
+P(n): 8.32 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 13731945.00 cycles
+P(n): 10.59 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 47216112.00 cycles
+P(n): 12.26 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 273902208.00 cycles
+P(n): 13.17 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1066031616.00 cycles
+P(n): 13.52 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 4168480512.00 cycles
+P(n): 13.83 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 25959776256.00 cycles
+P(n): 13.87 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 103256596480.00 cycles
+P(n): 5.63 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix/x86/3.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix/x86/3.log
new file mode 100644
index 0000000..a66e481
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix/x86/3.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4421310.50 cycles
+P(n): 8.31 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 13851305.00 cycles
+P(n): 10.50 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 47453300.00 cycles
+P(n): 12.20 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 275350432.00 cycles
+P(n): 13.10 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 1059425216.00 cycles
+P(n): 13.61 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 4161480448.00 cycles
+P(n): 13.85 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 25909958656.00 cycles
+P(n): 13.90 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 102891044864.00 cycles
+P(n): 5.65 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/1.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/1.log
new file mode 100644
index 0000000..8b8601d
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4116743.00 cycles
+P(n): 8.92 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14193060.00 cycles
+P(n): 10.25 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 46275016.00 cycles
+P(n): 12.51 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 250446432.00 cycles
+P(n): 14.40 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 977844992.00 cycles
+P(n): 14.74 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3811610624.00 cycles
+P(n): 15.12 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 23616888832.00 cycles
+P(n): 15.25 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 94000291840.00 cycles
+P(n): 6.18 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/2.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/2.log
new file mode 100644
index 0000000..2a209b9
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/2.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4130377.00 cycles
+P(n): 8.90 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14179700.00 cycles
+P(n): 10.26 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 46318220.00 cycles
+P(n): 12.50 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 250998336.00 cycles
+P(n): 14.37 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 982180672.00 cycles
+P(n): 14.68 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3805426944.00 cycles
+P(n): 15.14 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 23592302592.00 cycles
+P(n): 15.26 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 93937680384.00 cycles
+P(n): 6.19 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/3.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/3.log
new file mode 100644
index 0000000..7f8cb13
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/3.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4107580.00 cycles
+P(n): 8.94 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14142018.00 cycles
+P(n): 10.29 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 46132236.00 cycles
+P(n): 12.55 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 248790048.00 cycles
+P(n): 14.50 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 974417088.00 cycles
+P(n): 14.79 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3791391744.00 cycles
+P(n): 15.20 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 23488210944.00 cycles
+P(n): 15.33 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 93195214848.00 cycles
+P(n): 6.24 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix/x86/1.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix/x86/1.log
new file mode 100644
index 0000000..3ea961d
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4119329.50 cycles
+P(n): 8.92 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14140615.00 cycles
+P(n): 10.29 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 46104904.00 cycles
+P(n): 12.56 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 248777184.00 cycles
+P(n): 14.50 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 972575296.00 cycles
+P(n): 14.82 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3793779712.00 cycles
+P(n): 15.19 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 23662673920.00 cycles
+P(n): 15.22 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 93648879616.00 cycles
+P(n): 6.21 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix/x86/2.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix/x86/2.log
new file mode 100644
index 0000000..29375a1
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix/x86/2.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4131873.00 cycles
+P(n): 8.89 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14207365.00 cycles
+P(n): 10.24 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 46157304.00 cycles
+P(n): 12.54 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 249512304.00 cycles
+P(n): 14.46 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 974145152.00 cycles
+P(n): 14.80 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3796545280.00 cycles
+P(n): 15.18 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 23539767296.00 cycles
+P(n): 15.30 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 93385080832.00 cycles
+P(n): 6.22 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix/x86/3.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix/x86/3.log
new file mode 100644
index 0000000..d853e35
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix/x86/3.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4133901.25 cycles
+P(n): 8.89 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14286805.00 cycles
+P(n): 10.18 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 46514768.00 cycles
+P(n): 12.45 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 250415104.00 cycles
+P(n): 14.41 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 979476864.00 cycles
+P(n): 14.72 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3790536448.00 cycles
+P(n): 15.20 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 23435288576.00 cycles
+P(n): 15.36 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 93591674880.00 cycles
+P(n): 6.21 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix/x86/1.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix/x86/1.log
new file mode 100644
index 0000000..4022b05
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4119765.00 cycles
+P(n): 8.92 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14158772.00 cycles
+P(n): 10.27 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 46276376.00 cycles
+P(n): 12.51 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 249878624.00 cycles
+P(n): 14.44 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 982863872.00 cycles
+P(n): 14.67 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3826156544.00 cycles
+P(n): 15.06 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 23693496320.00 cycles
+P(n): 15.20 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 94222524416.00 cycles
+P(n): 6.17 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix/x86/2.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix/x86/2.log
new file mode 100644
index 0000000..4c1d8cf
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix/x86/2.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4102695.50 cycles
+P(n): 8.96 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14151872.00 cycles
+P(n): 10.28 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 46248272.00 cycles
+P(n): 12.52 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 249935680.00 cycles
+P(n): 14.43 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 977354240.00 cycles
+P(n): 14.75 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3821584896.00 cycles
+P(n): 15.08 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 23594674176.00 cycles
+P(n): 15.26 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 94093844480.00 cycles
+P(n): 6.18 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix/x86/3.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix/x86/3.log
new file mode 100644
index 0000000..a80e32c
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix/x86/3.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 4124268.00 cycles
+P(n): 8.91 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 14218675.00 cycles
+P(n): 10.23 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 46843476.00 cycles
+P(n): 12.36 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 250210640.00 cycles
+P(n): 14.42 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 982352128.00 cycles
+P(n): 14.67 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3823222016.00 cycles
+P(n): 15.07 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 23695175680.00 cycles
+P(n): 15.20 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 94304829440.00 cycles
+P(n): 6.16 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix/x86/1.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix/x86/1.log
new file mode 100644
index 0000000..fd8f66f
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 3472917.50 cycles
+P(n): 10.58 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 11763622.00 cycles
+P(n): 12.37 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 38070640.00 cycles
+P(n): 15.21 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 204154320.00 cycles
+P(n): 17.67 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 798999808.00 cycles
+P(n): 18.04 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3114275584.00 cycles
+P(n): 18.50 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 19146825728.00 cycles
+P(n): 18.81 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 76207964160.00 cycles
+P(n): 7.63 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix/x86/2.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix/x86/2.log
new file mode 100644
index 0000000..0a1f7d4
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix/x86/2.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 3503500.00 cycles
+P(n): 10.49 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 11802675.00 cycles
+P(n): 12.33 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 37947776.00 cycles
+P(n): 15.26 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 203323776.00 cycles
+P(n): 17.74 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 821986944.00 cycles
+P(n): 17.54 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3103710720.00 cycles
+P(n): 18.57 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 19254915072.00 cycles
+P(n): 18.70 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 76036898816.00 cycles
+P(n): 7.64 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix/x86/3.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix/x86/3.log
new file mode 100644
index 0000000..8fb271d
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix/x86/3.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 3474629.50 cycles
+P(n): 10.57 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 11869545.00 cycles
+P(n): 12.26 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 38112416.00 cycles
+P(n): 15.19 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 203449264.00 cycles
+P(n): 17.73 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 816738752.00 cycles
+P(n): 17.65 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3081998592.00 cycles
+P(n): 18.70 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 19241965568.00 cycles
+P(n): 18.71 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 76032835584.00 cycles
+P(n): 7.64 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved/x86/1.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved/x86/1.log
new file mode 100644
index 0000000..ed597a5
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 3443131.25 cycles
+P(n): 10.67 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 11289992.00 cycles
+P(n): 12.89 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 36293512.00 cycles
+P(n): 15.95 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 193259072.00 cycles
+P(n): 18.67 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 752713344.00 cycles
+P(n): 19.15 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 2940765440.00 cycles
+P(n): 19.60 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 18122862592.00 cycles
+P(n): 19.87 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 71869390848.00 cycles
+P(n): 8.09 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved/x86/2.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved/x86/2.log
new file mode 100644
index 0000000..08bfc3e
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved/x86/2.log
@@ -0,0 +1,64 @@
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 3387230.00 cycles
+P(n): 10.85 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 11326786.00 cycles
+P(n): 12.84 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 36351516.00 cycles
+P(n): 15.93 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 194490336.00 cycles
+P(n): 18.55 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 753201792.00 cycles
+P(n): 19.14 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 2954236416.00 cycles
+P(n): 19.51 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 18356828160.00 cycles
+P(n): 19.62 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 73678970880.00 cycles
+P(n): 7.89 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved/x86/3.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved/x86/3.log
new file mode 100644
index 0000000..70f3e89
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved/x86/3.log
@@ -0,0 +1,64 @@
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 3445767.00 cycles
+P(n): 10.66 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 11501729.00 cycles
+P(n): 12.65 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 36756200.00 cycles
+P(n): 15.75 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 198376320.00 cycles
+P(n): 18.18 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 765950336.00 cycles
+P(n): 18.82 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 2987576576.00 cycles
+P(n): 19.29 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 18625347584.00 cycles
+P(n): 19.33 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 73340780544.00 cycles
+P(n): 7.92 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp/x86/1.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp/x86/1.log
new file mode 100644
index 0000000..5b20e8e
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 2430659.75 cycles
+P(n): 15.12 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 8285106.00 cycles
+P(n): 17.56 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 30615050.00 cycles
+P(n): 18.91 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 186199040.00 cycles
+P(n): 19.37 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 728076288.00 cycles
+P(n): 19.80 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 2895743232.00 cycles
+P(n): 19.90 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 18180956160.00 cycles
+P(n): 19.81 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 72372396032.00 cycles
+P(n): 8.03 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp/x86/2.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp/x86/2.log
new file mode 100644
index 0000000..66eac2b
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp/x86/2.log
@@ -0,0 +1,64 @@
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 2443465.25 cycles
+P(n): 15.04 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 8370210.00 cycles
+P(n): 17.38 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 30839560.00 cycles
+P(n): 18.77 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 185338336.00 cycles
+P(n): 19.46 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 737704256.00 cycles
+P(n): 19.54 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 2915981056.00 cycles
+P(n): 19.76 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 18362775552.00 cycles
+P(n): 19.61 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 73707225088.00 cycles
+P(n): 7.88 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp/x86/3.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp/x86/3.log
new file mode 100644
index 0000000..a9c7550
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp/x86/3.log
@@ -0,0 +1,64 @@
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 2445340.00 cycles
+P(n): 15.02 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 8350645.00 cycles
+P(n): 17.42 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 30609110.00 cycles
+P(n): 18.91 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 186796768.00 cycles
+P(n): 19.31 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 733212800.00 cycles
+P(n): 19.66 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 2960800768.00 cycles
+P(n): 19.46 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 18505687040.00 cycles
+P(n): 19.46 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 73613836288.00 cycles
+P(n): 7.89 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix/x86/1.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix/x86/1.log
new file mode 100644
index 0000000..9eac607
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 3897775.00 cycles
+P(n): 9.43 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 13522278.00 cycles
+P(n): 10.76 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 50063060.00 cycles
+P(n): 11.56 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 240297888.00 cycles
+P(n): 15.01 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 942706688.00 cycles
+P(n): 15.29 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3722795008.00 cycles
+P(n): 15.48 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 22749288448.00 cycles
+P(n): 15.83 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 90110476288.00 cycles
+P(n): 6.45 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix/x86/2.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix/x86/2.log
new file mode 100644
index 0000000..e61cf75
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix/x86/2.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 3907767.50 cycles
+P(n): 9.40 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 13510660.00 cycles
+P(n): 10.77 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 50107380.00 cycles
+P(n): 11.55 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 239951200.00 cycles
+P(n): 15.03 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 940302272.00 cycles
+P(n): 15.33 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3722942720.00 cycles
+P(n): 15.48 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 22711234560.00 cycles
+P(n): 15.85 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 90064855040.00 cycles
+P(n): 6.45 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix/x86/3.log b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix/x86/3.log
new file mode 100644
index 0000000..98b4ec6
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix/x86/3.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float.cpp ./variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix.cpp -o main_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 56.42 flops/byte
+W(n): 36740000.00 flops
+T(n): 3864792.00 cycles
+P(n): 9.51 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 112.67 flops/byte
+W(n): 145480000.00 flops
+T(n): 13472545.00 cycles
+P(n): 10.80 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 225.17 flops/byte
+W(n): 578960000.00 flops
+T(n): 49757488.00 cycles
+P(n): 11.64 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 562.67 flops/byte
+W(n): 3607399936.00 flops
+T(n): 238792880.00 cycles
+P(n): 15.11 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 1125.17 flops/byte
+W(n): 14414799872.00 flops
+T(n): 953924224.00 cycles
+P(n): 15.11 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 2250.17 flops/byte
+W(n): 57629601792.00 flops
+T(n): 3701525248.00 cycles
+P(n): 15.57 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800369920.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 5625.17 flops/byte
+W(n): 360073986048.00 flops
+T(n): 22775025664.00 cycles
+P(n): 15.81 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772800.00 flops
+Q(n) per iteration: 640056 bytes
+I(n): 4539.87 flops/byte
+W(n): 581154570240.00 flops
+T(n): 90258751488.00 cycles
+P(n): 6.44 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_symmetry/x86/2.log b/measurements/variant_float_inlining_SoA_symmetry/x86/2.log
new file mode 100644
index 0000000..89a2fd1
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_symmetry/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry.cpp -o main_float_inlining_SoA_symmetry -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 20694936.00 cycles
+P(n): 0.92 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 81658352.00 cycles
+P(n): 0.91 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 334585664.00 cycles
+P(n): 0.87 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 2584741888.00 cycles
+P(n): 0.70 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 11100169216.00 cycles
+P(n): 0.65 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 49019121664.00 cycles
+P(n): 0.59 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 351723257856.00 cycles
+P(n): 0.51 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_symmetry/x86/3.log b/measurements/variant_float_inlining_SoA_symmetry/x86/3.log
new file mode 100644
index 0000000..38d9994
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_symmetry/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry.cpp -o main_float_inlining_SoA_symmetry -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 20725438.00 cycles
+P(n): 0.92 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 81754928.00 cycles
+P(n): 0.91 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 334743744.00 cycles
+P(n): 0.87 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 2581337088.00 cycles
+P(n): 0.70 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 11223846912.00 cycles
+P(n): 0.64 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 49533292544.00 cycles
+P(n): 0.58 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 350542561280.00 cycles
+P(n): 0.51 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_symmetry_ILP_16/x86/2.log b/measurements/variant_float_inlining_SoA_symmetry_ILP_16/x86/2.log
new file mode 100644
index 0000000..2967098
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_symmetry_ILP_16/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry_ILP_16.cpp -o main_float_inlining_SoA_symmetry_ILP_16 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 17928308.00 cycles
+P(n): 1.07 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 64051680.00 cycles
+P(n): 1.16 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 254104352.00 cycles
+P(n): 1.15 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 1873828992.00 cycles
+P(n): 0.97 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 8172222976.00 cycles
+P(n): 0.88 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 33889355776.00 cycles
+P(n): 0.85 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 218516439040.00 cycles
+P(n): 0.82 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_symmetry_ILP_16/x86/3.log b/measurements/variant_float_inlining_SoA_symmetry_ILP_16/x86/3.log
new file mode 100644
index 0000000..a185098
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_symmetry_ILP_16/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry_ILP_16.cpp -o main_float_inlining_SoA_symmetry_ILP_16 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 18233164.00 cycles
+P(n): 1.05 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 65275440.00 cycles
+P(n): 1.14 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 253765520.00 cycles
+P(n): 1.15 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 1865975808.00 cycles
+P(n): 0.97 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 8123003904.00 cycles
+P(n): 0.89 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 33680871424.00 cycles
+P(n): 0.86 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 220842590208.00 cycles
+P(n): 0.82 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_symmetry_ILP_2/x86/2.log b/measurements/variant_float_inlining_SoA_symmetry_ILP_2/x86/2.log
new file mode 100644
index 0000000..6ac08be
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_symmetry_ILP_2/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry_ILP_2.cpp -o main_float_inlining_SoA_symmetry_ILP_2 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 18636110.00 cycles
+P(n): 1.02 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 74353936.00 cycles
+P(n): 1.00 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 294626592.00 cycles
+P(n): 0.99 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 2190295296.00 cycles
+P(n): 0.83 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 9388432384.00 cycles
+P(n): 0.77 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 40763506688.00 cycles
+P(n): 0.71 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 288141017088.00 cycles
+P(n): 0.63 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_symmetry_ILP_2/x86/3.log b/measurements/variant_float_inlining_SoA_symmetry_ILP_2/x86/3.log
new file mode 100644
index 0000000..6e0974c
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_symmetry_ILP_2/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry_ILP_2.cpp -o main_float_inlining_SoA_symmetry_ILP_2 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 18660336.00 cycles
+P(n): 1.02 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 72993616.00 cycles
+P(n): 1.02 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 293862496.00 cycles
+P(n): 1.00 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 2185636352.00 cycles
+P(n): 0.83 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 9371756544.00 cycles
+P(n): 0.77 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 41083510784.00 cycles
+P(n): 0.70 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 287173902336.00 cycles
+P(n): 0.63 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_symmetry_ILP_4/x86/2.log b/measurements/variant_float_inlining_SoA_symmetry_ILP_4/x86/2.log
new file mode 100644
index 0000000..7faa71c
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_symmetry_ILP_4/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry_ILP_4.cpp -o main_float_inlining_SoA_symmetry_ILP_4 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 17093748.00 cycles
+P(n): 1.12 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 65814740.00 cycles
+P(n): 1.13 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 268615712.00 cycles
+P(n): 1.09 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 1957872256.00 cycles
+P(n): 0.92 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 8319636480.00 cycles
+P(n): 0.87 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 35145236480.00 cycles
+P(n): 0.82 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 237461798912.00 cycles
+P(n): 0.76 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_symmetry_ILP_4/x86/3.log b/measurements/variant_float_inlining_SoA_symmetry_ILP_4/x86/3.log
new file mode 100644
index 0000000..6d1d221
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_symmetry_ILP_4/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry_ILP_4.cpp -o main_float_inlining_SoA_symmetry_ILP_4 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 17023572.00 cycles
+P(n): 1.12 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 65836080.00 cycles
+P(n): 1.13 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 267259904.00 cycles
+P(n): 1.09 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 1952850944.00 cycles
+P(n): 0.93 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 8324440064.00 cycles
+P(n): 0.87 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 35432906752.00 cycles
+P(n): 0.81 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 238751940608.00 cycles
+P(n): 0.75 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_symmetry_ILP_8/x86/2.log b/measurements/variant_float_inlining_SoA_symmetry_ILP_8/x86/2.log
new file mode 100644
index 0000000..36b9d5c
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_symmetry_ILP_8/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry_ILP_8.cpp -o main_float_inlining_SoA_symmetry_ILP_8 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 15348080.00 cycles
+P(n): 1.24 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 56733160.00 cycles
+P(n): 1.31 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 226593312.00 cycles
+P(n): 1.29 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 1704302464.00 cycles
+P(n): 1.06 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 7344024576.00 cycles
+P(n): 0.98 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 31043047424.00 cycles
+P(n): 0.93 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 205332381696.00 cycles
+P(n): 0.88 flops/cycle
+
diff --git a/measurements/variant_float_inlining_SoA_symmetry_ILP_8/x86/3.log b/measurements/variant_float_inlining_SoA_symmetry_ILP_8/x86/3.log
new file mode 100644
index 0000000..222a7e0
--- /dev/null
+++ b/measurements/variant_float_inlining_SoA_symmetry_ILP_8/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-float_symmetry.cpp ./variants/variant_float_inlining_SoA_symmetry_ILP_8.cpp -o main_float_inlining_SoA_symmetry_ILP_8 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 3256 bytes
+I(n): 29.33 flops/byte
+W(n): 19100000.00 flops
+T(n): 15111212.00 cycles
+P(n): 1.26 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 6456 bytes
+I(n): 57.47 flops/byte
+W(n): 74200000.00 flops
+T(n): 55873808.00 cycles
+P(n): 1.33 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 12856 bytes
+I(n): 113.72 flops/byte
+W(n): 292400000.00 flops
+T(n): 225309888.00 cycles
+P(n): 1.30 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 32056 bytes
+I(n): 282.47 flops/byte
+W(n): 1811000064.00 flops
+T(n): 1706010624.00 cycles
+P(n): 1.06 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 64056 bytes
+I(n): 563.73 flops/byte
+W(n): 7222000128.00 flops
+T(n): 7389733376.00 cycles
+P(n): 0.98 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 128056 bytes
+I(n): 1126.23 flops/byte
+W(n): 28843999232.00 flops
+T(n): 31043899392.00 cycles
+P(n): 0.93 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550016.00 flops
+Q(n) per iteration: 320056 bytes
+I(n): 2813.73 flops/byte
+W(n): 180110000128.00 flops
+T(n): 205621460992.00 cycles
+P(n): 0.88 flops/cycle
+
diff --git a/measurements/variant_inlining_7/x86/2.log b/measurements/variant_inlining_7/x86/2.log
new file mode 100644
index 0000000..ecff983
--- /dev/null
+++ b/measurements/variant_inlining_7/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_7.cpp -o main_inlining_7 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 26033530.00 cycles
+P(n): 1.41 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 102659900.00 cycles
+P(n): 1.42 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 414240760.00 cycles
+P(n): 1.40 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 2701176340.00 cycles
+P(n): 1.34 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 10897753680.00 cycles
+P(n): 1.32 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 44787475960.00 cycles
+P(n): 1.29 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 302018381520.00 cycles
+P(n): 1.19 flops/cycle
+
diff --git a/measurements/variant_inlining_7/x86/3.log b/measurements/variant_inlining_7/x86/3.log
new file mode 100644
index 0000000..ea2a6b0
--- /dev/null
+++ b/measurements/variant_inlining_7/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_7.cpp -o main_inlining_7 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 26177315.00 cycles
+P(n): 1.40 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 103020520.00 cycles
+P(n): 1.41 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 414208700.00 cycles
+P(n): 1.40 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 2684751880.00 cycles
+P(n): 1.34 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 10888976120.00 cycles
+P(n): 1.32 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 44523656380.00 cycles
+P(n): 1.29 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 301445315720.00 cycles
+P(n): 1.19 flops/cycle
+
diff --git a/measurements/variant_inlining_7_SoA/x86/2.log b/measurements/variant_inlining_7_SoA/x86/2.log
new file mode 100644
index 0000000..5e198c7
--- /dev/null
+++ b/measurements/variant_inlining_7_SoA/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_7_SoA.cpp -o main_inlining_7_SoA -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 35265555.00 cycles
+P(n): 1.04 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 135241820.00 cycles
+P(n): 1.08 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 568722420.00 cycles
+P(n): 1.02 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 3531252960.00 cycles
+P(n): 1.02 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 13883478000.00 cycles
+P(n): 1.04 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 55083280220.00 cycles
+P(n): 1.05 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 339504049100.00 cycles
+P(n): 1.06 flops/cycle
+
diff --git a/measurements/variant_inlining_7_SoA/x86/3.log b/measurements/variant_inlining_7_SoA/x86/3.log
new file mode 100644
index 0000000..ee1aa07
--- /dev/null
+++ b/measurements/variant_inlining_7_SoA/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_7_SoA.cpp -o main_inlining_7_SoA -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 35315300.00 cycles
+P(n): 1.04 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 135382900.00 cycles
+P(n): 1.07 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 567506660.00 cycles
+P(n): 1.02 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 3508819820.00 cycles
+P(n): 1.03 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 13778499800.00 cycles
+P(n): 1.05 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 55289505680.00 cycles
+P(n): 1.04 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 339702161980.00 cycles
+P(n): 1.06 flops/cycle
+
diff --git a/measurements/variant_inlining_7_SoA_AVX2/x86/3.log b/measurements/variant_inlining_7_SoA_AVX2/x86/3.log
new file mode 100644
index 0000000..4303c26
--- /dev/null
+++ b/measurements/variant_inlining_7_SoA_AVX2/x86/3.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_7_SoA_AVX2.cpp -o main_inlining_7_SoA_AVX2 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 16889960.00 cycles
+P(n): 2.18 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 57609390.00 cycles
+P(n): 2.53 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 220531020.00 cycles
+P(n): 2.63 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 1352997020.00 cycles
+P(n): 2.67 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 5360580460.00 cycles
+P(n): 2.69 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 21347590360.00 cycles
+P(n): 2.70 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 119599550780.00 cycles
+P(n): 3.01 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 483808876720.00 cycles
+P(n): 1.20 flops/cycle
+
diff --git a/measurements/variant_inlining_7_SoA_AVX2_polyexp_expfix/x86/1.log b/measurements/variant_inlining_7_SoA_AVX2_polyexp_expfix/x86/1.log
new file mode 100644
index 0000000..7ff1112
--- /dev/null
+++ b/measurements/variant_inlining_7_SoA_AVX2_polyexp_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_7_SoA_AVX2_polyexp_expfix.cpp -o main_inlining_7_SoA_AVX2_polyexp_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 11371033.75 cycles
+P(n): 3.23 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 39890030.00 cycles
+P(n): 3.65 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 153171560.00 cycles
+P(n): 3.78 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 926036960.00 cycles
+P(n): 3.90 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 3676943300.00 cycles
+P(n): 3.92 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 14545721140.00 cycles
+P(n): 3.96 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 90702048200.00 cycles
+P(n): 3.97 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 362081083160.00 cycles
+P(n): 1.61 flops/cycle
+
diff --git a/measurements/variant_inlining_7_SoA_AVX2_polyexp_expfix/x86/2.log b/measurements/variant_inlining_7_SoA_AVX2_polyexp_expfix/x86/2.log
new file mode 100644
index 0000000..f46ef75
--- /dev/null
+++ b/measurements/variant_inlining_7_SoA_AVX2_polyexp_expfix/x86/2.log
@@ -0,0 +1,64 @@
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 11368170.00 cycles
+P(n): 3.23 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 39909820.00 cycles
+P(n): 3.65 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 152279380.00 cycles
+P(n): 3.80 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 925130680.00 cycles
+P(n): 3.90 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 3668254420.00 cycles
+P(n): 3.93 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 14660040260.00 cycles
+P(n): 3.93 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 93340296780.00 cycles
+P(n): 3.86 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 362367747120.00 cycles
+P(n): 1.60 flops/cycle
+
diff --git a/measurements/variant_inlining_7_SoA_AVX2_polyexp_expfix/x86/3.log b/measurements/variant_inlining_7_SoA_AVX2_polyexp_expfix/x86/3.log
new file mode 100644
index 0000000..25880a8
--- /dev/null
+++ b/measurements/variant_inlining_7_SoA_AVX2_polyexp_expfix/x86/3.log
@@ -0,0 +1,64 @@
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 11342471.25 cycles
+P(n): 3.24 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 40124180.00 cycles
+P(n): 3.63 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 151911440.00 cycles
+P(n): 3.81 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 923357200.00 cycles
+P(n): 3.91 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 3687447740.00 cycles
+P(n): 3.91 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 14732693600.00 cycles
+P(n): 3.91 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 90864696480.00 cycles
+P(n): 3.96 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 361983172200.00 cycles
+P(n): 1.61 flops/cycle
+
diff --git a/measurements/variant_inlining_7_symmetry_1/x86/2.log b/measurements/variant_inlining_7_symmetry_1/x86/2.log
new file mode 100644
index 0000000..92871fc
--- /dev/null
+++ b/measurements/variant_inlining_7_symmetry_1/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-symmetry.cpp ./variants/variant_inlining_7_symmetry_1.cpp -o main_inlining_7_symmetry_1 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 14.67 flops/byte
+W(n): 19100000.00 flops
+T(n): 26727245.00 cycles
+P(n): 0.71 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 28.73 flops/byte
+W(n): 74200000.00 flops
+T(n): 103992320.00 cycles
+P(n): 0.71 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 56.86 flops/byte
+W(n): 292400000.00 flops
+T(n): 420016640.00 cycles
+P(n): 0.70 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 141.24 flops/byte
+W(n): 1811000000.00 flops
+T(n): 2777078580.00 cycles
+P(n): 0.65 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 281.86 flops/byte
+W(n): 7222000000.00 flops
+T(n): 11331552620.00 cycles
+P(n): 0.64 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 563.11 flops/byte
+W(n): 28844000000.00 flops
+T(n): 47005533260.00 cycles
+P(n): 0.61 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 1406.86 flops/byte
+W(n): 180110000000.00 flops
+T(n): 207916013360.00 cycles
+P(n): 0.87 flops/cycle
+
diff --git a/measurements/variant_inlining_7_symmetry_1/x86/3.log b/measurements/variant_inlining_7_symmetry_1/x86/3.log
new file mode 100644
index 0000000..9f97be0
--- /dev/null
+++ b/measurements/variant_inlining_7_symmetry_1/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-symmetry.cpp ./variants/variant_inlining_7_symmetry_1.cpp -o main_inlining_7_symmetry_1 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 14.67 flops/byte
+W(n): 19100000.00 flops
+T(n): 26767850.00 cycles
+P(n): 0.71 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 28.73 flops/byte
+W(n): 74200000.00 flops
+T(n): 104659680.00 cycles
+P(n): 0.71 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 56.86 flops/byte
+W(n): 292400000.00 flops
+T(n): 421525360.00 cycles
+P(n): 0.69 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 141.24 flops/byte
+W(n): 1811000000.00 flops
+T(n): 2790761980.00 cycles
+P(n): 0.65 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 281.86 flops/byte
+W(n): 7222000000.00 flops
+T(n): 11641782400.00 cycles
+P(n): 0.62 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 563.11 flops/byte
+W(n): 28844000000.00 flops
+T(n): 47106595880.00 cycles
+P(n): 0.61 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 1406.86 flops/byte
+W(n): 180110000000.00 flops
+T(n): 208836888460.00 cycles
+P(n): 0.86 flops/cycle
+
diff --git a/measurements/variant_inlining_7_symmetry_2/x86/2.log b/measurements/variant_inlining_7_symmetry_2/x86/2.log
new file mode 100644
index 0000000..4b0911a
--- /dev/null
+++ b/measurements/variant_inlining_7_symmetry_2/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-symmetry.cpp ./variants/variant_inlining_7_symmetry_2.cpp -o main_inlining_7_symmetry_2 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 14.67 flops/byte
+W(n): 19100000.00 flops
+T(n): 26562245.00 cycles
+P(n): 0.72 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 28.73 flops/byte
+W(n): 74200000.00 flops
+T(n): 103075420.00 cycles
+P(n): 0.72 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 56.86 flops/byte
+W(n): 292400000.00 flops
+T(n): 418508180.00 cycles
+P(n): 0.70 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 141.24 flops/byte
+W(n): 1811000000.00 flops
+T(n): 2778344240.00 cycles
+P(n): 0.65 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 281.86 flops/byte
+W(n): 7222000000.00 flops
+T(n): 11360649140.00 cycles
+P(n): 0.64 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 563.11 flops/byte
+W(n): 28844000000.00 flops
+T(n): 46979623260.00 cycles
+P(n): 0.61 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 1406.86 flops/byte
+W(n): 180110000000.00 flops
+T(n): 208094812200.00 cycles
+P(n): 0.87 flops/cycle
+
diff --git a/measurements/variant_inlining_7_symmetry_2/x86/3.log b/measurements/variant_inlining_7_symmetry_2/x86/3.log
new file mode 100644
index 0000000..4cab342
--- /dev/null
+++ b/measurements/variant_inlining_7_symmetry_2/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-symmetry.cpp ./variants/variant_inlining_7_symmetry_2.cpp -o main_inlining_7_symmetry_2 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 14.67 flops/byte
+W(n): 19100000.00 flops
+T(n): 26620640.00 cycles
+P(n): 0.72 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 28.73 flops/byte
+W(n): 74200000.00 flops
+T(n): 103702040.00 cycles
+P(n): 0.72 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 56.86 flops/byte
+W(n): 292400000.00 flops
+T(n): 417450420.00 cycles
+P(n): 0.70 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 141.24 flops/byte
+W(n): 1811000000.00 flops
+T(n): 2785555280.00 cycles
+P(n): 0.65 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 281.86 flops/byte
+W(n): 7222000000.00 flops
+T(n): 11343649360.00 cycles
+P(n): 0.64 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 563.11 flops/byte
+W(n): 28844000000.00 flops
+T(n): 46953691640.00 cycles
+P(n): 0.61 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 1406.86 flops/byte
+W(n): 180110000000.00 flops
+T(n): 207442349960.00 cycles
+P(n): 0.87 flops/cycle
+
diff --git a/measurements/variant_inlining_7_symmetry_2_ilp_1/x86/2.log b/measurements/variant_inlining_7_symmetry_2_ilp_1/x86/2.log
new file mode 100644
index 0000000..979c643
--- /dev/null
+++ b/measurements/variant_inlining_7_symmetry_2_ilp_1/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-symmetry.cpp ./variants/variant_inlining_7_symmetry_2_ilp_1.cpp -o main_inlining_7_symmetry_2_ilp_1 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 14.67 flops/byte
+W(n): 19100000.00 flops
+T(n): 26434750.00 cycles
+P(n): 0.72 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 28.73 flops/byte
+W(n): 74200000.00 flops
+T(n): 103076140.00 cycles
+P(n): 0.72 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 56.86 flops/byte
+W(n): 292400000.00 flops
+T(n): 414752620.00 cycles
+P(n): 0.70 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 141.24 flops/byte
+W(n): 1811000000.00 flops
+T(n): 2727126060.00 cycles
+P(n): 0.66 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 281.86 flops/byte
+W(n): 7222000000.00 flops
+T(n): 11087200940.00 cycles
+P(n): 0.65 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 563.11 flops/byte
+W(n): 28844000000.00 flops
+T(n): 45051280280.00 cycles
+P(n): 0.64 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 1406.86 flops/byte
+W(n): 180110000000.00 flops
+T(n): 210209411220.00 cycles
+P(n): 0.86 flops/cycle
+
diff --git a/measurements/variant_inlining_7_symmetry_2_ilp_1/x86/3.log b/measurements/variant_inlining_7_symmetry_2_ilp_1/x86/3.log
new file mode 100644
index 0000000..d2f263a
--- /dev/null
+++ b/measurements/variant_inlining_7_symmetry_2_ilp_1/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-symmetry.cpp ./variants/variant_inlining_7_symmetry_2_ilp_1.cpp -o main_inlining_7_symmetry_2_ilp_1 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 14.67 flops/byte
+W(n): 19100000.00 flops
+T(n): 26387870.00 cycles
+P(n): 0.72 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 28.73 flops/byte
+W(n): 74200000.00 flops
+T(n): 102030060.00 cycles
+P(n): 0.73 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 56.86 flops/byte
+W(n): 292400000.00 flops
+T(n): 414172500.00 cycles
+P(n): 0.71 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 141.24 flops/byte
+W(n): 1811000000.00 flops
+T(n): 2720855140.00 cycles
+P(n): 0.67 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 281.86 flops/byte
+W(n): 7222000000.00 flops
+T(n): 11113087680.00 cycles
+P(n): 0.65 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 563.11 flops/byte
+W(n): 28844000000.00 flops
+T(n): 44960698300.00 cycles
+P(n): 0.64 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 1406.86 flops/byte
+W(n): 180110000000.00 flops
+T(n): 208987223120.00 cycles
+P(n): 0.86 flops/cycle
+
diff --git a/measurements/variant_inlining_7_symmetry_2_ilp_2/x86/2.log b/measurements/variant_inlining_7_symmetry_2_ilp_2/x86/2.log
new file mode 100644
index 0000000..260ce44
--- /dev/null
+++ b/measurements/variant_inlining_7_symmetry_2_ilp_2/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-symmetry.cpp ./variants/variant_inlining_7_symmetry_2_ilp_2.cpp -o main_inlining_7_symmetry_2_ilp_2 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 14.67 flops/byte
+W(n): 19100000.00 flops
+T(n): 26819620.00 cycles
+P(n): 0.71 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 28.73 flops/byte
+W(n): 74200000.00 flops
+T(n): 105029000.00 cycles
+P(n): 0.71 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 56.86 flops/byte
+W(n): 292400000.00 flops
+T(n): 423715960.00 cycles
+P(n): 0.69 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 141.24 flops/byte
+W(n): 1811000000.00 flops
+T(n): 2769934960.00 cycles
+P(n): 0.65 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 281.86 flops/byte
+W(n): 7222000000.00 flops
+T(n): 11885758320.00 cycles
+P(n): 0.61 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 563.11 flops/byte
+W(n): 28844000000.00 flops
+T(n): 48235104920.00 cycles
+P(n): 0.60 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 1406.86 flops/byte
+W(n): 180110000000.00 flops
+T(n): 317460685280.00 cycles
+P(n): 0.57 flops/cycle
+
diff --git a/measurements/variant_inlining_7_symmetry_2_ilp_2/x86/3.log b/measurements/variant_inlining_7_symmetry_2_ilp_2/x86/3.log
new file mode 100644
index 0000000..95972c7
--- /dev/null
+++ b/measurements/variant_inlining_7_symmetry_2_ilp_2/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main-symmetry.cpp ./variants/variant_inlining_7_symmetry_2_ilp_2.cpp -o main_inlining_7_symmetry_2_ilp_2 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 95500.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 14.67 flops/byte
+W(n): 19100000.00 flops
+T(n): 26682785.00 cycles
+P(n): 0.72 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 371000.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 28.73 flops/byte
+W(n): 74200000.00 flops
+T(n): 103925380.00 cycles
+P(n): 0.71 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 1462000.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 56.86 flops/byte
+W(n): 292400000.00 flops
+T(n): 415782740.00 cycles
+P(n): 0.70 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 9055000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 141.24 flops/byte
+W(n): 1811000000.00 flops
+T(n): 2735007900.00 cycles
+P(n): 0.66 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 36110000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 281.86 flops/byte
+W(n): 7222000000.00 flops
+T(n): 11240995220.00 cycles
+P(n): 0.64 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 144220000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 563.11 flops/byte
+W(n): 28844000000.00 flops
+T(n): 46790666220.00 cycles
+P(n): 0.62 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 900550000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 1406.86 flops/byte
+W(n): 180110000000.00 flops
+T(n): 308031751960.00 cycles
+P(n): 0.58 flops/cycle
+
diff --git a/measurements/variant_inlining_8/x86/2.log b/measurements/variant_inlining_8/x86/2.log
new file mode 100644
index 0000000..3a83c82
--- /dev/null
+++ b/measurements/variant_inlining_8/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8.cpp -o main_inlining_8 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 25924030.00 cycles
+P(n): 1.42 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 103751840.00 cycles
+P(n): 1.40 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 414349600.00 cycles
+P(n): 1.40 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 2738371940.00 cycles
+P(n): 1.32 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 10966457560.00 cycles
+P(n): 1.31 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 44694972240.00 cycles
+P(n): 1.29 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 303613890420.00 cycles
+P(n): 1.19 flops/cycle
+
diff --git a/measurements/variant_inlining_8/x86/3.log b/measurements/variant_inlining_8/x86/3.log
new file mode 100644
index 0000000..ff35d11
--- /dev/null
+++ b/measurements/variant_inlining_8/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8.cpp -o main_inlining_8 -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 25871680.00 cycles
+P(n): 1.42 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 101682420.00 cycles
+P(n): 1.43 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 411536620.00 cycles
+P(n): 1.41 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 2678464020.00 cycles
+P(n): 1.35 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 10860602260.00 cycles
+P(n): 1.33 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 44447633220.00 cycles
+P(n): 1.30 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 301546448660.00 cycles
+P(n): 1.19 flops/cycle
+
diff --git a/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/1.log b/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/1.log
new file mode 100644
index 0000000..b741ee1
--- /dev/null
+++ b/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp -o main_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 8339301.25 cycles
+P(n): 4.41 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 28662660.00 cycles
+P(n): 5.08 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 104402900.00 cycles
+P(n): 5.55 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 636519040.00 cycles
+P(n): 5.67 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 2499027620.00 cycles
+P(n): 5.77 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 9941452140.00 cycles
+P(n): 5.80 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 61650558000.00 cycles
+P(n): 5.84 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 246792591620.00 cycles
+P(n): 2.35 flops/cycle
+
diff --git a/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/2.log b/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/2.log
new file mode 100644
index 0000000..5762311
--- /dev/null
+++ b/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/2.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp -o main_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 8445907.50 cycles
+P(n): 4.35 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 28691310.00 cycles
+P(n): 5.07 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 104933160.00 cycles
+P(n): 5.52 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 639568720.00 cycles
+P(n): 5.64 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 2515964020.00 cycles
+P(n): 5.73 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 9996713720.00 cycles
+P(n): 5.76 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 62098746740.00 cycles
+P(n): 5.80 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 247013435860.00 cycles
+P(n): 2.35 flops/cycle
+
diff --git a/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/3.log b/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/3.log
new file mode 100644
index 0000000..f95f05f
--- /dev/null
+++ b/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix/x86/3.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp -o main_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 8335317.50 cycles
+P(n): 4.41 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 28557990.00 cycles
+P(n): 5.09 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 104404360.00 cycles
+P(n): 5.55 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 635858120.00 cycles
+P(n): 5.67 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 2509183840.00 cycles
+P(n): 5.74 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 9956726160.00 cycles
+P(n): 5.79 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 61839346220.00 cycles
+P(n): 5.82 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 246699025260.00 cycles
+P(n): 2.36 flops/cycle
+
diff --git a/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix/x86/1.log b/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix/x86/1.log
new file mode 100644
index 0000000..18b6f5f
--- /dev/null
+++ b/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix.cpp -o main_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 8308587.50 cycles
+P(n): 4.42 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 28404470.00 cycles
+P(n): 5.12 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 103978120.00 cycles
+P(n): 5.57 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 633362280.00 cycles
+P(n): 5.70 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 2494262100.00 cycles
+P(n): 5.78 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 9899453320.00 cycles
+P(n): 5.82 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 61865956520.00 cycles
+P(n): 5.82 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 246306063920.00 cycles
+P(n): 2.36 flops/cycle
+
diff --git a/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix/x86/2.log b/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix/x86/2.log
new file mode 100644
index 0000000..1653c69
--- /dev/null
+++ b/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix/x86/2.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix.cpp -o main_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 8344811.25 cycles
+P(n): 4.40 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 28520365.00 cycles
+P(n): 5.10 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 104404120.00 cycles
+P(n): 5.55 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 636400560.00 cycles
+P(n): 5.67 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 2501258800.00 cycles
+P(n): 5.76 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 9901533780.00 cycles
+P(n): 5.82 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 61815669960.00 cycles
+P(n): 5.82 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 247091597180.00 cycles
+P(n): 2.35 flops/cycle
+
diff --git a/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix/x86/3.log b/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix/x86/3.log
new file mode 100644
index 0000000..a60b0dc
--- /dev/null
+++ b/measurements/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix/x86/3.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix.cpp -o main_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 8321530.00 cycles
+P(n): 4.42 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 28458355.00 cycles
+P(n): 5.11 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 104209360.00 cycles
+P(n): 5.56 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 634333640.00 cycles
+P(n): 5.69 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 2496796020.00 cycles
+P(n): 5.77 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 9914787460.00 cycles
+P(n): 5.81 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 61721090840.00 cycles
+P(n): 5.83 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 246519066200.00 cycles
+P(n): 2.36 flops/cycle
+
diff --git a/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix/x86/1.log b/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix/x86/1.log
new file mode 100644
index 0000000..c33cf4f
--- /dev/null
+++ b/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix.cpp -o main_inlining_8_SoA_AVX512_polyexp_moreavx_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 7985528.75 cycles
+P(n): 4.60 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 27703355.00 cycles
+P(n): 5.25 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 102292560.00 cycles
+P(n): 5.66 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 629804220.00 cycles
+P(n): 5.73 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 2472263180.00 cycles
+P(n): 5.83 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 9851588140.00 cycles
+P(n): 5.85 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 61479499820.00 cycles
+P(n): 5.86 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 245617006060.00 cycles
+P(n): 2.37 flops/cycle
+
diff --git a/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix/x86/2.log b/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix/x86/2.log
new file mode 100644
index 0000000..5ee8171
--- /dev/null
+++ b/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix/x86/2.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix.cpp -o main_inlining_8_SoA_AVX512_polyexp_moreavx_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 8028216.25 cycles
+P(n): 4.58 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 27833730.00 cycles
+P(n): 5.23 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 102829680.00 cycles
+P(n): 5.63 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 629698060.00 cycles
+P(n): 5.73 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 2469782600.00 cycles
+P(n): 5.84 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 9826240720.00 cycles
+P(n): 5.86 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 61490138600.00 cycles
+P(n): 5.86 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 245848977420.00 cycles
+P(n): 2.36 flops/cycle
+
diff --git a/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix/x86/3.log b/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix/x86/3.log
new file mode 100644
index 0000000..27a1fc5
--- /dev/null
+++ b/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix/x86/3.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix.cpp -o main_inlining_8_SoA_AVX512_polyexp_moreavx_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 7992830.00 cycles
+P(n): 4.60 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 27693680.00 cycles
+P(n): 5.25 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 102340840.00 cycles
+P(n): 5.66 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 638328340.00 cycles
+P(n): 5.65 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 2468865800.00 cycles
+P(n): 5.84 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 9838254540.00 cycles
+P(n): 5.86 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 61344519940.00 cycles
+P(n): 5.87 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 245174501540.00 cycles
+P(n): 2.37 flops/cycle
+
diff --git a/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/1.log b/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/1.log
new file mode 100644
index 0000000..7a9a201
--- /dev/null
+++ b/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/1.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp -o main_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 8506121.25 cycles
+P(n): 4.32 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 27681145.00 cycles
+P(n): 5.26 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 98048030.00 cycles
+P(n): 5.90 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 594320700.00 cycles
+P(n): 6.07 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 2270890500.00 cycles
+P(n): 6.35 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 8985906260.00 cycles
+P(n): 6.41 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 56358587540.00 cycles
+P(n): 6.39 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 224065475420.00 cycles
+P(n): 2.59 flops/cycle
+
diff --git a/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/2.log b/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/2.log
new file mode 100644
index 0000000..e6443dc
--- /dev/null
+++ b/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/2.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp -o main_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 8512763.75 cycles
+P(n): 4.32 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 27679115.00 cycles
+P(n): 5.26 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 98109310.00 cycles
+P(n): 5.90 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 583896260.00 cycles
+P(n): 6.18 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 2292138700.00 cycles
+P(n): 6.29 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 9006037600.00 cycles
+P(n): 6.40 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 56211721500.00 cycles
+P(n): 6.41 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 223879818840.00 cycles
+P(n): 2.60 flops/cycle
+
diff --git a/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/3.log b/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/3.log
new file mode 100644
index 0000000..c702109
--- /dev/null
+++ b/measurements/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix/x86/3.log
@@ -0,0 +1,66 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp -o main_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 8550285.00 cycles
+P(n): 4.30 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 27834905.00 cycles
+P(n): 5.23 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 98660270.00 cycles
+P(n): 5.87 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 587210580.00 cycles
+P(n): 6.14 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 2280133880.00 cycles
+P(n): 6.32 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 8974668040.00 cycles
+P(n): 6.42 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 56052905260.00 cycles
+P(n): 6.42 flops/cycle
+
+=== N_P=10000 N_ITER=200 ===
+W(n) per iteration: 2905772704.00 flops
+Q(n) per iteration: 1280112 bytes
+I(n): 2269.94 flops/byte
+W(n): 581154540800.00 flops
+T(n): 224433219680.00 cycles
+P(n): 2.59 flops/cycle
+
diff --git a/measurements/variant_lut_exp/x86/2.log b/measurements/variant_lut_exp/x86/2.log
new file mode 100644
index 0000000..82f5137
--- /dev/null
+++ b/measurements/variant_lut_exp/x86/2.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_lut_exp.cpp -o main_lut_exp -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 28659980.00 cycles
+P(n): 1.28 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 108522660.00 cycles
+P(n): 1.34 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 435222760.00 cycles
+P(n): 1.33 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 2907689920.00 cycles
+P(n): 1.24 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 11907431920.00 cycles
+P(n): 1.21 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 47838796500.00 cycles
+P(n): 1.20 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 305274270720.00 cycles
+P(n): 1.18 flops/cycle
+
diff --git a/measurements/variant_lut_exp/x86/3.log b/measurements/variant_lut_exp/x86/3.log
new file mode 100644
index 0000000..bb31ac2
--- /dev/null
+++ b/measurements/variant_lut_exp/x86/3.log
@@ -0,0 +1,58 @@
+rm -f main_*  run_tests *.o *.s variants/*.s variants/*.o headers/*.s headers/*.o
+/usr/bin/g++ main.cpp ./variants/variant_lut_exp.cpp -o main_lut_exp -O3 -march=native -std=c++23 -Wall -Wextra -flto -DFLOPCOUNT=0 -DRANDOM_SEED=0 -DROOFLINE=0 -DBENCHMARK=1
+=== N_P=50 N_ITER=200 ===
+W(n) per iteration: 183700.00 flops
+Q(n) per iteration: 6512 bytes
+I(n): 28.21 flops/byte
+W(n): 36740000.00 flops
+T(n): 28650585.00 cycles
+P(n): 1.28 flops/cycle
+
+=== N_P=100 N_ITER=200 ===
+W(n) per iteration: 727400.00 flops
+Q(n) per iteration: 12912 bytes
+I(n): 56.34 flops/byte
+W(n): 145480000.00 flops
+T(n): 109151460.00 cycles
+P(n): 1.33 flops/cycle
+
+=== N_P=200 N_ITER=200 ===
+W(n) per iteration: 2894800.00 flops
+Q(n) per iteration: 25712 bytes
+I(n): 112.59 flops/byte
+W(n): 578960000.00 flops
+T(n): 436914960.00 cycles
+P(n): 1.33 flops/cycle
+
+=== N_P=500 N_ITER=200 ===
+W(n) per iteration: 18037000.00 flops
+Q(n) per iteration: 64112 bytes
+I(n): 281.34 flops/byte
+W(n): 3607400000.00 flops
+T(n): 2895243380.00 cycles
+P(n): 1.25 flops/cycle
+
+=== N_P=1000 N_ITER=200 ===
+W(n) per iteration: 72074000.00 flops
+Q(n) per iteration: 128112 bytes
+I(n): 562.59 flops/byte
+W(n): 14414800000.00 flops
+T(n): 11815027680.00 cycles
+P(n): 1.22 flops/cycle
+
+=== N_P=2000 N_ITER=200 ===
+W(n) per iteration: 288148000.00 flops
+Q(n) per iteration: 256112 bytes
+I(n): 1125.09 flops/byte
+W(n): 57629600000.00 flops
+T(n): 47796189940.00 cycles
+P(n): 1.21 flops/cycle
+
+=== N_P=5000 N_ITER=200 ===
+W(n) per iteration: 1800370000.00 flops
+Q(n) per iteration: 640112 bytes
+I(n): 2812.59 flops/byte
+W(n): 360074000000.00 flops
+T(n): 304323811480.00 cycles
+P(n): 1.18 flops/cycle
+
diff --git a/scripts/mem-bench.sh b/scripts/mem-bench.sh
index afce0e0..4e49d42 100755
--- a/scripts/mem-bench.sh
+++ b/scripts/mem-bench.sh
@@ -1 +1,11 @@
-sudo sysbench memory --memory-block-size=1G --memory-total-size=20G --memory-oper=read run
+#!/bin/bash
+
+ARCH="x86"
+if uname -m | grep -q "arm\|aarch64"; then
+	ARCH="arm"
+fi
+
+LOGICAL_CORE_COUNT=$(grep ^cpu\\scores /proc/cpuinfo | uniq |  awk '{print $4}')
+for run in {1..100}; do
+	sudo sysbench memory --memory-block-size=1G --memory-total-size=20G --memory-oper=read --threads=$LOGICAL_CORE_COUNT run > measurements/memory/$ARCH/$run.log
+done
diff --git a/variants/variant_AVX512_1_flopcount.cpp b/variants/variant_AVX512_1_flopcount.cpp
new file mode 100644
index 0000000..718ee9f
--- /dev/null
+++ b/variants/variant_AVX512_1_flopcount.cpp
@@ -0,0 +1,568 @@
+// # This file attempts to use AVX-512 intrinsics to speed up computation.
+// # Predecessor: variant_inlining_7.cpp
+#ifdef __AVX512F__
+
+#include "../headers/board_reduced.hpp"
+#include "../headers/constants.hpp"
+#include "../headers/pedestrian_reduced.hpp"
+
+#include "immintrin.h"
+#include "sleef.h"
+
+#include <algorithm>
+#include <cmath>
+#include <fstream>
+#include <immintrin.h>
+#include <iostream>
+#include <math.h>
+#include <random>
+#include <tuple>
+#include <vector>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0
+#define BOARD_LENGTH 50.0
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0f
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5f
+#define WALL_EPS 1e-3
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<double> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<double> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<double> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<double> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  Pedestrian *P = new Pedestrian[n_p];
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    double x = x_gen_left(gen);
+    double y = y_gen(gen);
+
+    double start_pos_x = x;
+    double start_pos_y = y;
+
+    double destination_x = BOARD_LENGTH + 30.0;
+    double destination_y = y;
+
+    double init_e_x = destination_x - start_pos_x;
+    double init_e_y = destination_y - start_pos_y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double desired_v0;
+    do {
+      desired_v0 = velocity_gen(gen);
+    } while (desired_v0 <= 0.1);
+
+    double v_max = 1.3 * desired_v0;
+
+    P[i] = Pedestrian(0, 0, 0, start_pos_x, start_pos_y, init_e_unit_x, init_e_unit_y, v_max, desired_v0, destination_x, destination_y);
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    double x = x_gen_right(gen);
+    double y = y_gen(gen);
+
+    double start_pos_x = x;
+    double start_pos_y = y;
+    double destination_x = -30;
+    double destination_y = y;
+
+    double init_e_x = destination_x - start_pos_x;
+    double init_e_y = destination_y - start_pos_y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double desired_v0;
+    do {
+      desired_v0 = velocity_gen(gen);
+    } while (desired_v0 <= 0.1);
+
+    double v_max = 1.3 * desired_v0;
+
+    P[i] = Pedestrian(1, 0, 0, start_pos_x, start_pos_y, init_e_unit_x, init_e_unit_y, v_max, desired_v0, destination_x, destination_y);
+  }
+
+  return P;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+
+  vector<tuple<int, int, double, double, double, double, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+      samples.emplace_back(0, i, p.curr_pos_x, p.curr_pos_y, p.e_x, p.e_y, p.color);
+    }
+  }
+#endif
+
+  // # ADDs:  n_iter * n_p * 4 +
+  // #        n_iter * n_p * ceil(((double) (n_p - 7)) / 8) * (1 / (ceil(((double) (n_p - 7)) / 8))) * 7 * 14 +
+  // #        n_iter * n_p * ceil(((double) (n_p - 7)) / 8) * (1 - 1 / (ceil(((double) (n_p - 7)) / 8))) * 224 +
+  // #        n_iter * n_p * (n_p - floor((double) n_p / 8) * 8) * 14 +
+  // #        n_iter * n_p * board.n_obstacles * 12 +
+  // #        n_iter * n_p * 7
+
+  // # MULTs: n_iter * n_p * 4 +
+  // #        n_iter * n_p * ceil(((double) (n_p - 7)) / 8) * (1 / (ceil(((double) (n_p - 7)) / 8))) * 7 * 24 +
+  // #        n_iter * n_p * ceil(((double) (n_p - 7)) / 8) * (1 - 1 / (ceil(((double) (n_p - 7)) / 8))) * 352 +
+  // #        n_iter * n_p * (n_p - floor((double) n_p / 8) * 8) * 24 +
+  // #        n_iter * n_p * board.n_obstacles * 19 +
+  // #        n_iter * n_p * 17
+  
+  // # DIVs:  n_iter * n_p * ceil(((double) (n_p - 7)) / 8) * (1 / (ceil(((double) (n_p - 7)) / 8))) * 7 +
+  // #        n_iter * n_p * ceil(((double) (n_p - 7)) / 8) * (1 - 1 / (ceil(((double) (n_p - 7)) / 8))) * 8 +
+  // #        n_iter * n_p * (n_p - floor((double) n_p / 8) * 8) +
+  // #        n_iter * n_p * board.n_obstacles +
+  // #        n_iter * n_p * 2
+  
+  // # SQRTs: n_iter * n_p * ceil(((double) (n_p - 7)) / 8) * (1 / (ceil(((double) (n_p - 7)) / 8))) * 7 * 3 +
+  // #        n_iter * n_p * ceil(((double) (n_p - 7)) / 8) * (1 - 1 / (ceil(((double) (n_p - 7)) / 8))) * 24 +
+  // #        n_iter * n_p * (n_p - floor((double) n_p / 8) * 8) * 3 +
+  // #        n_iter * n_p * board.n_obstacles +
+  // #        n_iter * n_p * 3
+
+  // # EXPs:  n_iter * n_p * ceil(((double) (n_p - 7)) / 8) * (1 / (ceil(((double) (n_p - 7)) / 8))) * 7 +
+  // #        n_iter * n_p * (n_p - floor((double) n_p / 8) * 8) +
+  // #        n_iter * n_p * board.n_obstacles
+
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+
+      // # 2 MULTs
+      double desired_velocity_x = p.e_x * p.desired_v0;
+      double desired_velocity_y = p.e_y * p.desired_v0;
+
+      // # 2 ADDs, 2 MULTs
+      double f_alpha_dt_x = (desired_velocity_x - p.curr_vel_x) * dt_times_tau_alpha_inv;
+      double f_alpha_dt_y = (desired_velocity_y - p.curr_vel_y) * dt_times_tau_alpha_inv;
+
+      __m512d desired_velocity_x_vec = _mm512_set1_pd(desired_velocity_x);
+      __m512d desired_velocity_y_vec = _mm512_set1_pd(desired_velocity_y);
+      __m512d f_alpha_dt_x_vec = _mm512_set1_pd(f_alpha_dt_x);
+      __m512d f_alpha_dt_y_vec = _mm512_set1_pd(f_alpha_dt_y);
+
+      __m512d alpha_curr_pos_x_vec = _mm512_set1_pd(p.curr_pos_x);
+      __m512d alpha_curr_pos_y_vec = _mm512_set1_pd(p.curr_pos_y);
+      __m512d alpha_e_x_vec = _mm512_set1_pd(p.e_x);
+      __m512d alpha_e_y_vec = _mm512_set1_pd(p.e_y);
+
+      int j;
+
+      // # ############################################################################################################################################################################################
+
+      for (j = 0; j < n_p - 7; j += 8) {
+
+        if (j <= i && i <= j + 7) {
+
+          for (int jj = j; jj < j + 8; jj++) {
+
+            if (jj == i) {
+              continue;
+            }
+
+            // # 2 ADDs
+            double r_ab_x = p.curr_pos_x - pedestrians[jj].curr_pos_x;
+            double r_ab_y = p.curr_pos_y - pedestrians[jj].curr_pos_y;
+
+            // # 1 SQRT, 2 MULTs, 1 ADD
+            double r_ab_norm = sqrt(r_ab_x * r_ab_x + r_ab_y * r_ab_y);
+
+            // # 2 ADDs
+            double r_ab2_x = r_ab_x - pedestrians[jj].s_b_e_x;
+            double r_ab2_y = r_ab_y - pedestrians[jj].s_b_e_y;
+
+            // # 1 SQRT, 2 MULTs, 1 ADD
+            double r_ab2_norm = sqrt(r_ab2_x * r_ab2_x + r_ab2_y * r_ab2_y);
+
+            // # 2 ADDs, 1 MULT
+            double r_ab_norm_sum = r_ab_norm + r_ab2_norm;
+            double two_b_sq = r_ab_norm_sum * r_ab_norm_sum - pedestrians[jj].s_b_sq;
+
+            // # 1 SQRT
+            double two_b = sqrt(two_b_sq);
+
+            // # 5 MULTs, 1 DIV, 1 EXP
+            double scalar_mult_common = (r_ab_norm_sum / (two_b * r_ab_norm * r_ab2_norm)) * (NEG_V0_AB_div_TWO_SIGMA * exp(two_b * NEG_TWO_SIGMA_INV));
+
+            // # 2 MULTs
+            double scalar_mult_r_ab = scalar_mult_common * r_ab2_norm;
+            double scalar_mult_r_ab2 = scalar_mult_common * r_ab_norm;
+
+            // # 4 MULTs, 2 ADDs
+            double f_alpha_beta_neg_x = r_ab_x * scalar_mult_r_ab + r_ab2_x * scalar_mult_r_ab2;
+            double f_alpha_beta_neg_y = r_ab_y * scalar_mult_r_ab + r_ab2_y * scalar_mult_r_ab2;
+
+            double w_val;
+
+            // # 3 MULTs, 1 ADD
+            double a = (p.e_x * f_alpha_beta_neg_x + p.e_y * f_alpha_beta_neg_y) * cos_phi_inv;
+
+            // # 2 MULTs, 1 ADD
+            double b_sq = f_alpha_beta_neg_x * f_alpha_beta_neg_x + f_alpha_beta_neg_y * f_alpha_beta_neg_y;
+
+            w_val = dt_neg;
+            // # 1 MULT
+            if (a >= 0 && (a * a > b_sq)) {
+              w_val = c_neg_times_dt;
+            }
+
+            // # 2 MULTs, 2 ADDs
+            f_alpha_dt_x = f_alpha_dt_x + f_alpha_beta_neg_x * w_val;
+            f_alpha_dt_y = f_alpha_dt_y + f_alpha_beta_neg_y * w_val;
+          }
+
+        } else {
+          __m512d beta_curr_pos_x_vec = _mm512_setr_pd(pedestrians[j].curr_pos_x, pedestrians[j + 1].curr_pos_x, pedestrians[j + 2].curr_pos_x, pedestrians[j + 3].curr_pos_x, pedestrians[j + 4].curr_pos_x, pedestrians[j + 5].curr_pos_x, pedestrians[j + 6].curr_pos_x, pedestrians[j + 7].curr_pos_x);
+          __m512d beta_curr_pos_y_vec = _mm512_setr_pd(pedestrians[j].curr_pos_y, pedestrians[j + 1].curr_pos_y, pedestrians[j + 2].curr_pos_y, pedestrians[j + 3].curr_pos_y, pedestrians[j + 4].curr_pos_y, pedestrians[j + 5].curr_pos_y, pedestrians[j + 6].curr_pos_y, pedestrians[j + 7].curr_pos_y);
+
+          // # 16 ADDs
+          __m512d r_ab_x_vec = _mm512_sub_pd(alpha_curr_pos_x_vec, beta_curr_pos_x_vec);
+          __m512d r_ab_y_vec = _mm512_sub_pd(alpha_curr_pos_y_vec, beta_curr_pos_y_vec);
+
+          // # 8 MULTs
+          __m512d r_ab_y_mul_r_ab_y_vec = _mm512_mul_pd(r_ab_y_vec, r_ab_y_vec);
+
+          // # 8 MULTs, 8 ADDs
+          __m512d r_ab_norm_sq_vec = _mm512_fmadd_pd(r_ab_x_vec, r_ab_x_vec, r_ab_y_mul_r_ab_y_vec);
+
+          // # 8 SQRTs
+          __m512d r_ab_norm_vec = _mm512_sqrt_pd(r_ab_norm_sq_vec);
+
+          __m512d beta_s_b_e_x_vec = _mm512_setr_pd(pedestrians[j].s_b_e_x, pedestrians[j + 1].s_b_e_x, pedestrians[j + 2].s_b_e_x, pedestrians[j + 3].s_b_e_x, pedestrians[j + 4].s_b_e_x, pedestrians[j + 5].s_b_e_x, pedestrians[j + 6].s_b_e_x, pedestrians[j + 7].s_b_e_x);
+
+          // # 8 ADDs
+          __m512d r_ab2_x_vec = _mm512_sub_pd(r_ab_x_vec, beta_s_b_e_x_vec);
+
+          __m512d beta_s_b_e_y_vec = _mm512_setr_pd(pedestrians[j].s_b_e_y, pedestrians[j + 1].s_b_e_y, pedestrians[j + 2].s_b_e_y, pedestrians[j + 3].s_b_e_y, pedestrians[j + 4].s_b_e_y, pedestrians[j + 5].s_b_e_y, pedestrians[j + 6].s_b_e_y, pedestrians[j + 7].s_b_e_y);
+
+          // # 8 ADDs
+          __m512d r_ab2_y_vec = _mm512_sub_pd(r_ab_y_vec, beta_s_b_e_y_vec);
+
+          // # 8 MULTs
+          __m512d r_ab2_y_mul_r_ab2_y_vec = _mm512_mul_pd(r_ab2_y_vec, r_ab2_y_vec);
+
+          // # 8 MULTs, 8 ADDs
+          __m512d r_ab2_norm_sq_vec = _mm512_fmadd_pd(r_ab2_x_vec, r_ab2_x_vec, r_ab2_y_mul_r_ab2_y_vec);
+
+          // # 8 SQRTs
+          __m512d r_ab2_norm_vec = _mm512_sqrt_pd(r_ab2_norm_sq_vec);
+
+          // # 8 ADDs
+          __m512d r_ab_norm_sum_vec = _mm512_add_pd(r_ab_norm_vec, r_ab2_norm_vec);
+
+          __m512d beta_s_b_sq_vec = _mm512_setr_pd(pedestrians[j].s_b_sq, pedestrians[j + 1].s_b_sq, pedestrians[j + 2].s_b_sq, pedestrians[j + 3].s_b_sq, pedestrians[j + 4].s_b_sq, pedestrians[j + 5].s_b_sq, pedestrians[j + 6].s_b_sq, pedestrians[j + 7].s_b_sq);
+
+          // # 8 MULTs, 8 ADDs
+          __m512d two_b_sq_vec = _mm512_fmsub_pd(r_ab_norm_sum_vec, r_ab_norm_sum_vec, beta_s_b_sq_vec);
+
+          // # 8 SQRTs
+          __m512d two_b_vec = _mm512_sqrt_pd(two_b_sq_vec);
+
+          // # 8 MULTs
+          __m512d two_b_mul_r_ab_norm_vec = _mm512_mul_pd(two_b_vec, r_ab_norm_vec);
+
+          // # 8 MULTs
+          __m512d denom_vec = _mm512_mul_pd(two_b_mul_r_ab_norm_vec, r_ab2_norm_vec);
+
+          // # 8 DIVs
+          __m512d scalar_mult_common_left_vec = _mm512_div_pd(r_ab_norm_sum_vec, denom_vec);
+
+          // # 8 MULTs
+          __m512d two_b_mul_NEG_TWO_SIGMA_INV_vec = _mm512_mul_pd(two_b_vec, NEG_TWO_SIGMA_INV_VEC);
+
+          // # 160 MULTs, 112 ADDs
+          // objdump -d --demangle /usr/lib/x86_64-linux-gnu/libsleef.so | grep -A 67 "Sleef_expd8_u10@@Base"
+          __m512d e_vec = Sleef_expd8_u10(two_b_mul_NEG_TWO_SIGMA_INV_vec);
+
+          // # 16 MULTs
+          __m512d scalar_mult_common_right_vec = _mm512_mul_pd(NEG_V0_AB_DIV_TWO_SIGMA_VEC, e_vec);
+          __m512d scalar_mult_common_vec = _mm512_mul_pd(scalar_mult_common_left_vec, scalar_mult_common_right_vec);
+
+          // # 16 MULTs
+          __m512d scalar_mult_r_ab_vec = _mm512_mul_pd(scalar_mult_common_vec, r_ab2_norm_vec);
+          __m512d scalar_mult_r_ab2_vec = _mm512_mul_pd(scalar_mult_common_vec, r_ab_norm_vec);
+
+          // # 16 MULTs
+          __m512d r_ab2_x_mul_scalar_mult_r_ab2_vec = _mm512_mul_pd(r_ab2_x_vec, scalar_mult_r_ab2_vec);
+          __m512d r_ab2_y_mul_scalar_mult_r_ab2_vec = _mm512_mul_pd(r_ab2_y_vec, scalar_mult_r_ab2_vec);
+
+          // # 16 MULTs, 16 ADDs
+          __m512d f_alpha_beta_neg_x_vec = _mm512_fmadd_pd(r_ab_x_vec, scalar_mult_r_ab_vec, r_ab2_x_mul_scalar_mult_r_ab2_vec);
+          __m512d f_alpha_beta_neg_y_vec = _mm512_fmadd_pd(r_ab_y_vec, scalar_mult_r_ab_vec, r_ab2_y_mul_scalar_mult_r_ab2_vec);
+
+          // # 8 MULTs
+          __m512d rgt_vec = _mm512_mul_pd(alpha_e_y_vec, f_alpha_beta_neg_y_vec);
+
+          // # 8 MULTs, 8 ADDs
+          __m512d lft_add_rgt_vec = _mm512_fmadd_pd(alpha_e_x_vec, f_alpha_beta_neg_x_vec, rgt_vec);
+
+          // # 8 MULTs
+          __m512d a_vec = _mm512_mul_pd(lft_add_rgt_vec, cos_phi_inv_vec);
+
+          // # 8 MULTs
+          __m512d f_alpha_beta_neg_y_sq_vec = _mm512_mul_pd(f_alpha_beta_neg_y_vec, f_alpha_beta_neg_y_vec);
+
+          // # 8 MULTs, 8 ADDs
+          __m512d b_sq_vec = _mm512_fmadd_pd(f_alpha_beta_neg_x_vec, f_alpha_beta_neg_x_vec, f_alpha_beta_neg_y_sq_vec);
+
+          __m512d w_val_vec = dt_neg_vec;
+          __mmask8 a_ge_zero_mask = _mm512_cmp_pd_mask(a_vec, ZERO_VEC, _CMP_GE_OQ);
+
+          // # 8 MULTs
+          __m512d a_sq_vec = _mm512_mul_pd(a_vec, a_vec);
+
+          __mmask8 a_sq_gt_b_sq_mask = _mm512_cmp_pd_mask(a_sq_vec, b_sq_vec, _CMP_GT_OQ);
+          __mmask8 cond_mask = _kand_mask8(a_ge_zero_mask, a_sq_gt_b_sq_mask);
+          w_val_vec = _mm512_mask_mov_pd(w_val_vec, cond_mask, c_neg_times_dt_vec);
+
+          // # 16 MULTs
+          __m512d x_term_vec = _mm512_mul_pd(f_alpha_beta_neg_x_vec, w_val_vec);
+          __m512d y_term_vec = _mm512_mul_pd(f_alpha_beta_neg_y_vec, w_val_vec);
+
+          // # 16 ADDs
+          f_alpha_dt_x += _mm512_reduce_add_pd(x_term_vec);
+          f_alpha_dt_y += _mm512_reduce_add_pd(y_term_vec);
+        }
+      }
+
+      // # ############################################################################################################################################################################################
+
+      for (; j < n_p; j++) {
+
+        if (j == i) {
+          continue;
+        }
+
+        // # 2 ADDs
+        double r_ab_x = p.curr_pos_x - pedestrians[j].curr_pos_x;
+        double r_ab_y = p.curr_pos_y - pedestrians[j].curr_pos_y;
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        double r_ab_norm = sqrt(r_ab_x * r_ab_x + r_ab_y * r_ab_y);
+
+        // # 2 ADDs
+        double r_ab2_x = r_ab_x - pedestrians[j].s_b_e_x;
+        double r_ab2_y = r_ab_y - pedestrians[j].s_b_e_y;
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        double r_ab2_norm = sqrt(r_ab2_x * r_ab2_x + r_ab2_y * r_ab2_y);
+
+        // # 1 ADD
+        double r_ab_norm_sum = r_ab_norm + r_ab2_norm;
+
+        // # 1 MULT, 1 ADD
+        double two_b_sq = r_ab_norm_sum * r_ab_norm_sum - pedestrians[j].s_b_sq;
+
+        // # 1 SQRT
+        double two_b = sqrt(two_b_sq);
+
+        // # 5 MULTs, 1 DIV, 1 EXP
+        double scalar_mult_common = (r_ab_norm_sum / (two_b * r_ab_norm * r_ab2_norm)) * (NEG_V0_AB_div_TWO_SIGMA * exp(two_b * NEG_TWO_SIGMA_INV));
+
+        // # 2 MULTs
+        double scalar_mult_r_ab = scalar_mult_common * r_ab2_norm;
+        double scalar_mult_r_ab2 = scalar_mult_common * r_ab_norm;
+
+        // # 4 MULTs, 2 ADDs
+        double f_alpha_beta_neg_x = r_ab_x * scalar_mult_r_ab + r_ab2_x * scalar_mult_r_ab2;
+        double f_alpha_beta_neg_y = r_ab_y * scalar_mult_r_ab + r_ab2_y * scalar_mult_r_ab2;
+
+        double w_val;
+
+        // # 3 MULTs, 1 ADD
+        double a = (p.e_x * f_alpha_beta_neg_x + p.e_y * f_alpha_beta_neg_y) * cos_phi_inv;
+
+        // # 2 MULTs, 1 ADD
+        double b_sq = f_alpha_beta_neg_x * f_alpha_beta_neg_x + f_alpha_beta_neg_y * f_alpha_beta_neg_y;
+
+        w_val = dt_neg;
+        // # 1 MULT
+        if (a >= 0 && (a * a > b_sq)) {
+          w_val = c_neg_times_dt;
+        }
+
+        // # 2 MULTs, 2 ADDs
+        f_alpha_dt_x = f_alpha_dt_x + f_alpha_beta_neg_x * w_val;
+        f_alpha_dt_y = f_alpha_dt_y + f_alpha_beta_neg_y * w_val;
+      }
+
+      // # ############################################################################################################################################################################################
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
+        double v1_x = p.curr_pos_x - board.obstacles[k].start_x;
+        double v1_y = p.curr_pos_y - board.obstacles[k].start_y;
+
+        // # 3 MULTs, 1 ADD
+        double projection = (v1_x * board.obstacles[k].obstacle_v_x + v1_y * board.obstacles[k].obstacle_v_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0)
+          projection = 0.0;
+        else if (projection > 1.0)
+          projection = 1.0;
+
+        // # 2 MULTs, 4 ADDs
+        double r_aB_x = p.curr_pos_x - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        double r_aB_y = p.curr_pos_y - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        double r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+
+        // # 2 MULTs, 1 EXP, 1 DIV
+        double scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+
+        // # 2 MULTs
+        double f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        double f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        double w_val;
+
+        // # 3 MULTs, 1 ADD
+        double a = (p.e_x * f_alpha_B_neg_x + p.e_y * f_alpha_B_neg_y) * cos_phi_inv;
+
+        // # 2 MULTs, 1 ADD
+        double b_sq = f_alpha_B_neg_x * f_alpha_B_neg_x + f_alpha_B_neg_y * f_alpha_B_neg_y;
+
+        w_val = dt_neg;
+
+        // # 1 MULT
+        if (a >= 0 && (a * a > b_sq)) {
+          w_val = c_neg_times_dt;
+        }
+
+        // # 2 MULTs
+        double f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        double f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        // # 2 ADDs
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      // # 2 ADDs
+      p.w_x += f_alpha_dt_x;
+      p.w_y += f_alpha_dt_y;
+    }
+
+    // # ############################################################################################################################################################################################
+
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+
+      // # 1 SQRT, 2 MULTs, 1 ADD
+      double w_norm = sqrt(p.w_x * p.w_x + p.w_y * p.w_y);
+
+      // # 1 DIV
+      double correction_factor = min(1.0, p.v_max / w_norm);
+
+      // # 2 MULTs
+      p.curr_vel_x = p.w_x * correction_factor;
+      p.curr_vel_y = p.w_y * correction_factor;
+
+      // # 2 MULTs, 2 ADDs
+      p.curr_pos_x = p.curr_pos_x + p.curr_vel_x * dt;
+      p.curr_pos_y = p.curr_pos_y + p.curr_vel_y * dt;
+
+      if (p.color == 0 && p.curr_pos_x > BOARD_LENGTH_MINUS_ONE_POINT_FIVE) {
+        p.curr_pos_x = 1.0;
+        p.destination_y = p.curr_pos_y;
+        p.w_x = p.desired_v0;
+        p.w_y = 0.0;
+        p.curr_vel_x = p.w_x;
+        p.curr_vel_y = p.w_y;
+      }
+      if (p.color == 1 && p.curr_pos_x < 1.5) {
+        p.curr_pos_x = BOARD_LENGTH_MINUS_ONE;
+        p.destination_y = p.curr_pos_y;
+        p.w_x = -p.desired_v0;
+        p.w_y = 0.0;
+        p.curr_vel_x = p.w_x;
+        p.curr_vel_y = p.w_y;
+      }
+
+      p.curr_pos_y = clamp(p.curr_pos_y, WALL_EPS, BOARD_WIDTH_MINUS_WALL_EPS);
+
+      // # 2 ADDs
+      double e_x = p.destination_x - p.curr_pos_x;
+      double e_y = p.destination_y - p.curr_pos_y;
+
+      // # 1 DIV, 1 SQRT, 2 MULTs, 1 ADD
+      double e_norm_inv = 1.0 / sqrt(e_x * e_x + e_y * e_y);
+
+      // # 2 MULTs
+      p.e_x = e_x * e_norm_inv;
+      p.e_y = e_y * e_norm_inv;
+
+      // # 2 MULTs, 1 ADD
+      double velocity_norm_sq = p.curr_vel_x * p.curr_vel_x + p.curr_vel_y * p.curr_vel_y;
+
+      // # 1 SQRT
+      double velocity_norm = sqrt(velocity_norm_sq);
+
+      // # 5 MULTs
+      p.s_b_sq = velocity_norm_sq * delta_t_sq;
+      p.s_b_e_x = p.e_x * (velocity_norm * delta_t);
+      p.s_b_e_y = p.e_y * (velocity_norm * delta_t);
+
+#if !BENCHMARK
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, p.curr_pos_x, p.curr_pos_y, p.e_x, p.e_y, p.color);
+      }
+#endif
+    }
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
+
+#endif
diff --git a/variants/variant_baseline_flopcount.cpp b/variants/variant_baseline_flopcount.cpp
new file mode 100644
index 0000000..6f4f90e
--- /dev/null
+++ b/variants/variant_baseline_flopcount.cpp
@@ -0,0 +1,440 @@
+// This file is merged from the forces.cpp and social_forces_baseline.cpp
+// The goal is that we from now on only have one file as our baseline
+// This should make it more easy in the future to implement optimizations
+// Our naming convention from now on is: variant_<variant_name>.cpp (Please make sure to keep that consistent, otherwise our Makefile wont work anymore)
+
+#include "../headers/board.hpp"
+#include "../headers/constants.hpp"
+#include "../headers/pedestrian.hpp"
+#include "../headers/vector2d.hpp"
+
+#include <algorithm>
+#include <cmath>
+#include <fstream>
+#include <iostream>
+#include <math.h> // keep — forces code uses unqualified sqrt/exp
+#include <random>
+#include <tuple>
+#include <vector>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0
+#define BOARD_LENGTH 50.0
+#define WALL_EPS 1e-3
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+// forces.cpp
+
+// Flop count: 2 ADDs, 6 MULTs
+double w_ped(Vector2d e, Vector2d f) {
+  double f_norm_sq = f.dot(f);
+  double rhs = f_norm_sq * cos_phi_sq;
+  double e_dot_f = e.dot(f);
+  double e_dot_f_sq = e_dot_f * e_dot_f;
+  if (e_dot_f >= 0.0 || e_dot_f_sq <= rhs)
+    return dt_neg;
+  else
+    return c_neg_times_dt;
+}
+
+double w_obstacle(Vector2d e, Vector2d f) {
+  double e_dot_f = (f.y == 0) ? 0 : (f.y < 0 ? -e.y : e.y);
+  if (e_dot_f >= cos_phi)
+    return dt_neg;
+  else
+    return c_neg_times_dt;
+}
+
+// Flop count: 2 ADDs, 4 MULTs
+Vector2d acceleration_term_dt(const Pedestrian &alpha) {
+
+  // # 2 MULTs
+  Vector2d desired_velocity = alpha.e * alpha.desired_v0;
+
+  Vector2d actual_velocity = alpha.curr_vel;
+
+  // # 2 ADDs, 2 MULTs
+  Vector2d f_alpha_0_dt = (desired_velocity - actual_velocity) * dt_times_tau_alpha_inv;
+
+  return f_alpha_0_dt;
+}
+
+// Flop count: 12 ADDs, 24 MULTs, 1 DIV, 3 SQRTs, 1 EXP
+Vector2d pedestrian_repulsion_dt(const Pedestrian &alpha, const Pedestrian &beta) {
+
+  // # 2 ADDs
+  // $\vec{r}_{\alpha\beta}$
+  Vector2d r_ab = alpha.curr_pos - beta.curr_pos;
+
+  // # 1 ADD, 2 MULTs, 1 SQRT
+  // $\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert$
+  double r_ab_norm = r_ab.norm();
+  r_ab_norm = r_ab_norm > EPS ? r_ab_norm : EPS; // Avoid division by zero
+
+  // $v_\beta \Delta t \vec{e}_\beta$
+  Vector2d s_b_e = beta.s_b_e;
+
+  // # 2 ADDs
+  // $\vec{r}_{\alpha\beta} - \vec{v}_{\beta}\varDelta_{t}\vec{e}_{\beta}$
+  Vector2d r_ab2 = r_ab - s_b_e;
+
+  // # 1 ADD, 2 MULTs, 1 SQRT
+  // $\Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert$
+  double r_ab2_norm = r_ab2.norm();
+  r_ab2_norm = r_ab2_norm > EPS ? r_ab2_norm : EPS; // Avoid division by zero
+
+  // # 1 ADD
+  // $\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert$
+  double r_ab_norm_sum = r_ab_norm + r_ab2_norm;
+
+  // # 1 ADD, 1 MULTs, 1 SQRT
+  //
+  // $2b = \sqrt{(\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert)^2 - (v_\beta \Delta t)^2}$
+  //
+  double two_b_sq = r_ab_norm_sum * r_ab_norm_sum - beta.s_b_sq;
+  two_b_sq = two_b_sq > EPS_SQ ? two_b_sq : EPS_SQ; // Avoid sqrt of negative and avoid division by zero
+  double two_b = sqrt(two_b_sq);
+
+  // # 5 MULTs, 1 DIV, 1 EXP
+  //
+  // $\frac{-V^0_{\alpha\beta}}{2\sigma}\,e^{-b/\sigma}\,\frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert\,\Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}$
+  //
+  double scalar_mult_common = (r_ab_norm_sum / (two_b * r_ab_norm * r_ab2_norm)) * (NEG_V0_AB_div_TWO_SIGMA * exp(two_b * NEG_TWO_SIGMA_INV));
+
+  // # 1 MULT
+  //
+  // $\frac{-V^0_{\alpha\beta}}{2\sigma}\,e^{-b/\sigma}\,\frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert}$
+  //
+  double scalar_mult_r_ab = scalar_mult_common * r_ab2_norm;
+
+  // # 1 MULT
+  //
+  // $\frac{-V^0_{\alpha\beta}}{2\sigma}\,e^{-b/\sigma}\,\frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}$
+  //
+  double scalar_mult_r_ab2 = scalar_mult_common * r_ab_norm;
+
+  // # 2 ADDs, 4 MULTs
+  //
+  // $\vec{-f}_{\alpha \beta}(\vec{r}_{\alpha \beta}) = \frac{-V^0_{\alpha \beta}}{2\sigma}e^{-b / \sigma} 2\grad_{\vec{r}_{\alpha \beta}}b $
+  //
+  //
+  //          $= \frac{-V^0_{\alpha\beta}}{2\sigma}\,e^{-b/\sigma}\,\frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert}\,\vec{r}_{\alpha \beta} + \frac{-V^0_{\alpha\beta}}{2\sigma}\,e^{-b/\sigma}\,\frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} -
+  //          v_\beta \Delta t \vec{e}_\beta \Arrowvert}\,(\vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta)$
+  //
+  //
+  //
+  // $2\grad_{\vec{r}_{\alpha \beta}}b = \frac{(\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert)(\frac{\vec{r}_{\alpha \beta}}{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert} + \frac{\vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta}{\Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert})}{2b}$
+  //
+  //
+  //       $= \frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert}\,\vec{r}_{\alpha \beta} + \frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}\,(\vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta)$
+  //
+  //
+  Vector2d f_alpha_beta_neg = r_ab * scalar_mult_r_ab + r_ab2 * scalar_mult_r_ab2;
+
+  // # 2 ADDs, 8 MULTs
+  //
+  // $\vec{F}_{\alpha\beta}(\vec{e}_{\alpha}, \vec{r}_{\alpha \beta}).dt = w(\vec{e}_{\alpha}, -\vec{f}_{\alpha\beta}(\vec{r}_{\alpha \beta})).\vec{f}_{\alpha\beta}(\vec{r}_{\alpha \beta}).dt$
+  //
+  Vector2d f_alpha_beta_dt = f_alpha_beta_neg * w_ped(alpha.e, f_alpha_beta_neg);
+
+  return f_alpha_beta_dt;
+}
+
+// Flop count: 8 ADDs, 13 MULTs, 1 DIV, 1 SQRT, 1 EXP
+Vector2d obstacle_repulsion_dt(const Pedestrian &alpha, const Obstacle &B) {
+
+  // # 7 ADDs, 5 MULTs
+  // $\vec{r}_{\alpha B}$
+  Vector2d r_aB = alpha.curr_pos - B.find_nearest_obstacle_point(alpha.curr_pos);
+
+  // # 1 ADD, 2 MULTs, 1 SQRT
+  // $\Arrowvert \vec{r}_{\alpha B} \Arrowvert$
+  double r_aB_norm = r_aB.norm();
+  r_aB_norm = r_aB_norm > EPS ? r_aB_norm : EPS; // Avoid division by zero
+
+  // # 2 MULTs, 1 DIV, 1 EXP
+  //
+  // $\frac{-U^0_{\alpha B}}{R} e^{- \Arrowvert \vec{r}_{\alpha B} \Arrowvert / R}$
+  //
+  double scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+
+  // # 2 MULTs
+  //
+  // $-\vec{F}_{\alpha B}(\vec{r}_{\alpha B}) = \frac{-U^0_{\alpha B}}{R} e^{- \Arrowvert \vec{r}_{\alpha B} \Arrowvert / R} \frac{\vec{r}_{\alpha B}}{\Arrowvert \vec{r}_{\alpha B} \Arrowvert}$
+  //
+  Vector2d f_alpha_B_neg = r_aB * scalar_mult;
+
+  // # 2 MULTs
+  //
+  // $\vec{F}_{\alpha B}(\vec{e}_{\alpha}, \vec{r}_{\alpha B}).dt = w(\vec{e}_{\alpha}, -\vec{F}_{\alpha B}(\vec{r}_{\alpha B})).\vec{F}_{\alpha B}(\vec{r}_{\alpha B}).dt$
+  //
+  Vector2d f_alpha_B_dt = f_alpha_B_neg * w_obstacle(alpha.e, f_alpha_B_neg);
+
+  return f_alpha_B_dt;
+}
+
+// social_force_baseline.cpp
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;  // Platform-provided random seed source.
+  mt19937 gen(rd()); // Non-reproducible initialization.
+#else
+  mt19937 gen(SEED); // Reproducible initialization.
+#endif
+
+  normal_distribution<double> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<double> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<double> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<double> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  Pedestrian *P = new Pedestrian[n_p];
+
+  int n_p_half = n_p / 2;
+
+  // Group of pedestrians walking from left to right.
+  for (int i = 0; i < n_p_half; i++) {
+    double x = x_gen_left(gen);
+    double y = y_gen(gen);
+
+    Vector2d start_pos = Vector2d(x, y);
+    Vector2d destination = Vector2d(BOARD_LENGTH + 30.0, y);
+
+    Vector2d init_e = destination - start_pos;
+    double init_e_norm = init_e.norm();
+    Vector2d init_e_unit = init_e / init_e_norm;
+
+    // Reject near-zero desired speeds to avoid pedestrians that barely move,
+    // and reject negative speeds because they would reverse the intended
+    // walking direction.
+    double desired_v0;
+    do {
+      desired_v0 = velocity_gen(gen);
+    } while (desired_v0 <= 0.1);
+
+    double v_max = 1.3 * desired_v0;
+
+    P[i] = Pedestrian(WHITE, Vector2d(), start_pos, init_e_unit, v_max, desired_v0, destination);
+  }
+
+  // Group of pedestrians walking from right to left.
+  for (int i = n_p_half; i < n_p; i++) {
+    double x = x_gen_right(gen);
+    double y = y_gen(gen);
+
+    Vector2d start_pos = Vector2d(x, y);
+    Vector2d destination = Vector2d(-30.0, y);
+
+    Vector2d init_e = destination - start_pos;
+    double init_e_norm = init_e.norm();
+    Vector2d init_e_unit = init_e / init_e_norm;
+
+    // Reject near-zero desired speeds to avoid pedestrians that barely move,
+    // and reject negative speeds because they would reverse the intended
+    // walking direction.
+    double desired_v0;
+    do {
+      desired_v0 = velocity_gen(gen);
+    } while (desired_v0 <= 0.1);
+
+    double v_max = 1.3 * desired_v0;
+
+    P[i] = Pedestrian(BLACK, Vector2d(), start_pos, init_e_unit, v_max, desired_v0, destination);
+  }
+
+  return P;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+
+#if BENCHMARK
+  // Avoid compiler unused parameter warning
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+
+  // Each sample stores: time step, pedestrian id, x, y, e.x, e.y, color.
+  vector<tuple<int, int, double, double, double, double, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+
+    // Store the initial state before the first simulation step.
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+      samples.emplace_back(0, i, p.curr_pos.x, p.curr_pos.y, p.e.x, p.e.y, p.color);
+    }
+  }
+#endif
+
+  // # ADDs: n_iter * (
+  // #   n_p * 4 +
+  // #   n_p * (n_p - 1) * 14 +
+  // #   n_p * board.n_obstacles * 10 +
+  // #   n_p * 7
+  // # )
+
+  // # MULTs: n_iter * (
+  // #   n_p * 4
+  // #   n_p * (n_p - 1) * 24 +
+  // #   n_p * board.n_obstacles * 13 +
+  // #   n_p * 16
+  // # )
+
+  // # DIVs: n_iter * (
+  // #   n_p * (n_p - 1) +
+  // #   n_p * board.n_obstacles +
+  // #   n_p * 2
+  // # )
+
+  // # SQRTs: n_iter * (
+  // #   n_p * (n_p - 1) * 3 +
+  // #   n_p * board.n_obstacles +
+  // #   n_p * 3
+  // # )
+
+  // # EXPs: n_iter * (
+  // #   n_p * (n_p - 1) +
+  // #   n_p * board.n_obstacles
+  // # )
+
+  for (int t = 0; t < n_iter; t++) {
+    // Pass 1: Compute all forces from the current state.
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+
+      // # 2 ADDs, 4 MULTs
+      //
+      // $\vec{F}_\alpha(\vec{v}_\alpha, v^0_\alpha \vec{e}_\alpha) = \frac{1}{\tau_\alpha}(v^0_\alpha \vec{e}_\alpha - \vec{v}_\alpha)$
+      //
+      Vector2d f_alpha_dt = acceleration_term_dt(p);
+
+      // # (N-1) * (14 ADDs, 24 MULTs, 1 DIV, 3 SQRTs, 1 EXP)
+      //
+      // $\sum_{\beta} \vec{F}_{\alpha\beta}(\vec{e}_{\alpha},\vec{r}_{\alpha} - \vec{r}_{\beta})$
+      //
+      for (int j = 0; j < n_p; j++) {
+        if (i == j)
+          continue;
+
+        f_alpha_dt += pedestrian_repulsion_dt(p, pedestrians[j]);
+      }
+
+      // # board.n_obstacles * (10 ADDs, 13 MULTs, 1 DIV, 1 SQRT, 1 EXP)
+      //
+      // $\sum_{B} \vec{F}_{\alpha B}(\vec{e}_{\alpha}, \vec{r}_{\alpha} - \vec{r}_{B})$
+      //
+      for (int k = 0; k < board.n_obstacles; k++) {
+        f_alpha_dt += obstacle_repulsion_dt(p, board.obstacles[k]);
+      }
+
+      // # 2 ADDs
+      //
+      //
+      // $\frac{d\vec{w}_\alpha}{dt} = \vec{F}_{\alpha}(t)$
+      //
+      p.w += f_alpha_dt;
+    }
+
+    // Pass 2: Update pedestrian states after all forces have been computed.
+    // # (7 ADDs, 16 MULTs, 2 DIVs, 3 SQRTs) * n_p
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+
+      // # 1 ADD, 2 MULTs, 1 SQRT
+      double w_norm = p.w.norm();
+
+      double w_norm2 = w_norm > EPS ? w_norm : EPS; // Avoid division by zero.
+
+      // # 1 DIV
+      double correction_factor = std::min(1.0, p.v_max / w_norm2);
+
+      // # 2 MULTs
+      p.curr_vel = p.w * correction_factor;
+
+      // # 2 ADDs, 2 MULTs
+      //
+      // $\vec{r}_i^{t+1} = \vec{r}_i^t + \vec{v}_i^t dt$
+      //
+      p.curr_pos = p.curr_pos + p.curr_vel * dt;
+
+      // When a pedestrian reaches the end of the corridor, teleport them back
+      // to the opposite entrance while keeping the same y-coordinate. This
+      // maintains a constant pedestrian flow and makes lane formation easier to
+      // observe.
+      if (p.color == WHITE && p.curr_pos.x > (BOARD_LENGTH - 1.5)) {
+        p.curr_pos.x = 1.0;
+        p.destination.y = p.curr_pos.y;
+        p.w = Vector2d(p.desired_v0, 0.0);
+        p.curr_vel = p.w;
+      }
+      if (p.color == BLACK && p.curr_pos.x < 1.5) {
+        p.curr_pos.x = BOARD_LENGTH - 1.0;
+        p.destination.y = p.curr_pos.y;
+        p.w = Vector2d(-p.desired_v0, 0.0);
+        p.curr_vel = p.w;
+      }
+
+      // Keep pedestrians inside the board boundaries.
+      p.curr_pos.y = clamp(p.curr_pos.y, WALL_EPS, BOARD_WIDTH_MINUS_WALL_EPS);
+
+      // # 2 ADDs
+      Vector2d e = p.destination - p.curr_pos;
+
+      // # 1 ADD, 2 MULTs, 1 DIV, 1 SQRT
+      double e_norm = e.norm();
+      double e_norm2 = e_norm > EPS ? e_norm : EPS; // Avoid division by zero.
+      double e_norm_inv = 1.0 / e_norm2;
+
+      // # 2 MULTs
+      p.e = e * e_norm_inv;
+
+      // # 1 ADD, 2 MULTs
+      double velocity_norm_sq = p.curr_vel.dot(p.curr_vel);
+
+      // # 1 SQRT
+      double velocity_norm = std::sqrt(velocity_norm_sq);
+
+      // # 1 MULT
+      p.s_b_sq = velocity_norm_sq * delta_t_sq;
+
+      // # 3 MULTs
+      p.s_b_e = p.e * (velocity_norm * delta_t);
+
+#if !BENCHMARK
+      // Store the current pedestrian state for visualization.
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, p.curr_pos.x, p.curr_pos.y, p.e.x, p.e.y, p.color);
+      }
+#endif
+    }
+  }
+
+#if !BENCHMARK
+  // Export recorded simulation states to CSV.
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
\ No newline at end of file
diff --git a/variants/variant_float_baseline_flopcount.cpp b/variants/variant_float_baseline_flopcount.cpp
new file mode 100644
index 0000000..820f769
--- /dev/null
+++ b/variants/variant_float_baseline_flopcount.cpp
@@ -0,0 +1,438 @@
+// # Changes: Added per-instruction flop count
+// # Predecessor: variant_float_baseline.cpp
+
+#include "../headers/board_float.hpp"
+#include "../headers/constants_float.hpp"
+#include "../headers/pedestrian_float.hpp"
+#include "../headers/vector2d_float.hpp"
+
+#include <algorithm>
+#include <cmath>
+#include <fstream>
+#include <iostream>
+#include <math.h> // keep — forces code uses unqualified sqrt/exp
+#include <random>
+#include <tuple>
+#include <vector>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0f
+#define BOARD_LENGTH 50.0f
+#define WALL_EPS 1e-3f
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999f
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+// forces.cpp
+
+// Flop count: 2 ADDs, 6 MULTs
+float w_ped(Vector2d e, Vector2d f) {
+  float f_norm_sq = f.dot(f);
+  float rhs = f_norm_sq * cos_phi_sq;
+  float e_dot_f = e.dot(f);
+  float e_dot_f_sq = e_dot_f * e_dot_f;
+  if (e_dot_f >= 0.0 || e_dot_f_sq <= rhs)
+    return dt_neg;
+  else
+    return c_neg_times_dt;
+}
+
+float w_obstacle(Vector2d e, Vector2d f) {
+  float e_dot_f = (f.y == 0) ? 0 : (f.y < 0 ? -e.y : e.y);
+  if (e_dot_f >= cos_phi)
+    return dt_neg;
+  else
+    return c_neg_times_dt;
+}
+
+// Flop count: 2 ADDs, 4 MULTs
+Vector2d acceleration_term_dt(const Pedestrian &alpha) {
+
+  // # 2 MULTs
+  Vector2d desired_velocity = alpha.e * alpha.desired_v0;
+
+  Vector2d actual_velocity = alpha.curr_vel;
+
+  // # 2 ADDs, 2 MULTs
+  Vector2d f_alpha_0_dt = (desired_velocity - actual_velocity) * dt_times_tau_alpha_inv;
+
+  return f_alpha_0_dt;
+}
+
+// Flop count: 12 ADDs, 24 MULTs, 1 DIV, 3 SQRTs, 1 EXP
+Vector2d pedestrian_repulsion_dt(const Pedestrian &alpha, const Pedestrian &beta) {
+
+  // # 2 ADDs
+  // $\vec{r}_{\alpha\beta}$
+  Vector2d r_ab = alpha.curr_pos - beta.curr_pos;
+
+  // # 1 ADD, 2 MULTs, 1 SQRT
+  // $\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert$
+  float r_ab_norm = r_ab.norm();
+  r_ab_norm = r_ab_norm > EPS ? r_ab_norm : EPS; // Avoid division by zero
+
+  // $v_\beta \Delta t \vec{e}_\beta$
+  Vector2d s_b_e = beta.s_b_e;
+
+  // # 2 ADDs
+  // $\vec{r}_{\alpha\beta} - \vec{v}_{\beta}\varDelta_{t}\vec{e}_{\beta}$
+  Vector2d r_ab2 = r_ab - s_b_e;
+
+  // # 1 ADD, 2 MULTs, 1 SQRT
+  // $\Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert$
+  float r_ab2_norm = r_ab2.norm();
+  r_ab2_norm = r_ab2_norm > EPS ? r_ab2_norm : EPS; // Avoid division by zero
+
+  // # 1 ADD
+  // $\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert$
+  float r_ab_norm_sum = r_ab_norm + r_ab2_norm;
+
+  // # 1 ADD, 1 MULTs, 1 SQRT
+  //
+  // $2b = \sqrt{(\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert)^2 - (v_\beta \Delta t)^2}$
+  //
+  float two_b_sq = r_ab_norm_sum * r_ab_norm_sum - beta.s_b_sq;
+  two_b_sq = two_b_sq > EPS_SQ ? two_b_sq : EPS_SQ; // Avoid sqrt of negative and avoid division by zero
+  float two_b = sqrt(two_b_sq);
+
+  // # 5 MULTs, 1 DIV, 1 EXP
+  //
+  // $\frac{-V^0_{\alpha\beta}}{2\sigma}\,e^{-b/\sigma}\,\frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert\,\Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}$
+  //
+  float scalar_mult_common = (r_ab_norm_sum / (two_b * r_ab_norm * r_ab2_norm)) * (NEG_V0_AB_div_TWO_SIGMA * exp(two_b * NEG_TWO_SIGMA_INV));
+
+  // # 1 MULT
+  //
+  // $\frac{-V^0_{\alpha\beta}}{2\sigma}\,e^{-b/\sigma}\,\frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert}$
+  //
+  float scalar_mult_r_ab = scalar_mult_common * r_ab2_norm;
+
+  // # 1 MULT
+  //
+  // $\frac{-V^0_{\alpha\beta}}{2\sigma}\,e^{-b/\sigma}\,\frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}$
+  //
+  float scalar_mult_r_ab2 = scalar_mult_common * r_ab_norm;
+
+  // # 2 ADDs, 4 MULTs
+  //
+  // $\vec{-f}_{\alpha \beta}(\vec{r}_{\alpha \beta}) = \frac{-V^0_{\alpha \beta}}{2\sigma}e^{-b / \sigma} 2\grad_{\vec{r}_{\alpha \beta}}b $
+  //
+  //
+  //          $= \frac{-V^0_{\alpha\beta}}{2\sigma}\,e^{-b/\sigma}\,\frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert}\,\vec{r}_{\alpha \beta} + \frac{-V^0_{\alpha\beta}}{2\sigma}\,e^{-b/\sigma}\,\frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} -
+  //          v_\beta \Delta t \vec{e}_\beta \Arrowvert}\,(\vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta)$
+  //
+  //
+  //
+  // $2\grad_{\vec{r}_{\alpha \beta}}b = \frac{(\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert)(\frac{\vec{r}_{\alpha \beta}}{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert} + \frac{\vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta}{\Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert})}{2b}$
+  //
+  //
+  //       $= \frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert}\,\vec{r}_{\alpha \beta} + \frac{\Arrowvert \vec{r}_{\alpha \beta} \Arrowvert + \Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}{2b\,\Arrowvert \vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta \Arrowvert}\,(\vec{r}_{\alpha \beta} - v_\beta \Delta t \vec{e}_\beta)$
+  //
+  //
+  Vector2d f_alpha_beta_neg = r_ab * scalar_mult_r_ab + r_ab2 * scalar_mult_r_ab2;
+
+  // # 2 ADDs, 8 MULTs
+  //
+  // $\vec{F}_{\alpha\beta}(\vec{e}_{\alpha}, \vec{r}_{\alpha \beta}).dt = w(\vec{e}_{\alpha}, -\vec{f}_{\alpha\beta}(\vec{r}_{\alpha \beta})).\vec{f}_{\alpha\beta}(\vec{r}_{\alpha \beta}).dt$
+  //
+  Vector2d f_alpha_beta_dt = f_alpha_beta_neg * w_ped(alpha.e, f_alpha_beta_neg);
+
+  return f_alpha_beta_dt;
+}
+
+// Flop count: 8 ADDs, 13 MULTs, 1 DIV, 1 SQRT, 1 EXP
+Vector2d obstacle_repulsion_dt(const Pedestrian &alpha, const Obstacle &B) {
+
+  // # 7 ADDs, 5 MULTs
+  // $\vec{r}_{\alpha B}$
+  Vector2d r_aB = alpha.curr_pos - B.find_nearest_obstacle_point(alpha.curr_pos);
+
+  // # 1 ADD, 2 MULTs, 1 SQRT
+  // $\Arrowvert \vec{r}_{\alpha B} \Arrowvert$
+  float r_aB_norm = r_aB.norm();
+  r_aB_norm = r_aB_norm > EPS ? r_aB_norm : EPS; // Avoid division by zero
+
+  // # 2 MULTs, 1 DIV, 1 EXP
+  //
+  // $\frac{-U^0_{\alpha B}}{R} e^{- \Arrowvert \vec{r}_{\alpha B} \Arrowvert / R}$
+  //
+  float scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+
+  // # 2 MULTs
+  //
+  // $-\vec{F}_{\alpha B}(\vec{r}_{\alpha B}) = \frac{-U^0_{\alpha B}}{R} e^{- \Arrowvert \vec{r}_{\alpha B} \Arrowvert / R} \frac{\vec{r}_{\alpha B}}{\Arrowvert \vec{r}_{\alpha B} \Arrowvert}$
+  //
+  Vector2d f_alpha_B_neg = r_aB * scalar_mult;
+
+  // # 2 MULTs
+  //
+  // $\vec{F}_{\alpha B}(\vec{e}_{\alpha}, \vec{r}_{\alpha B}).dt = w(\vec{e}_{\alpha}, -\vec{F}_{\alpha B}(\vec{r}_{\alpha B})).\vec{F}_{\alpha B}(\vec{r}_{\alpha B}).dt$
+  //
+  Vector2d f_alpha_B_dt = f_alpha_B_neg * w_obstacle(alpha.e, f_alpha_B_neg);
+
+  return f_alpha_B_dt;
+}
+
+// social_force_baseline.cpp
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;  // Platform-provided random seed source.
+  mt19937 gen(rd()); // Non-reproducible initialization.
+#else
+  mt19937 gen(SEED); // Reproducible initialization.
+#endif
+
+  normal_distribution<float> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<float> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<float> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<float> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  Pedestrian *P = new Pedestrian[n_p];
+
+  int n_p_half = n_p / 2;
+
+  // Group of pedestrians walking from left to right.
+  for (int i = 0; i < n_p_half; i++) {
+    float x = x_gen_left(gen);
+    float y = y_gen(gen);
+
+    Vector2d start_pos = Vector2d(x, y);
+    Vector2d destination = Vector2d(BOARD_LENGTH + 30.0, y);
+
+    Vector2d init_e = destination - start_pos;
+    float init_e_norm = init_e.norm();
+    Vector2d init_e_unit = init_e / init_e_norm;
+
+    // Reject near-zero desired speeds to avoid pedestrians that barely move,
+    // and reject negative speeds because they would reverse the intended
+    // walking direction.
+    float desired_v0;
+    do {
+      desired_v0 = velocity_gen(gen);
+    } while (desired_v0 <= 0.1);
+
+    float v_max = 1.3 * desired_v0;
+
+    P[i] = Pedestrian(WHITE, Vector2d(), start_pos, init_e_unit, v_max, desired_v0, destination);
+  }
+
+  // Group of pedestrians walking from right to left.
+  for (int i = n_p_half; i < n_p; i++) {
+    float x = x_gen_right(gen);
+    float y = y_gen(gen);
+
+    Vector2d start_pos = Vector2d(x, y);
+    Vector2d destination = Vector2d(-30.0, y);
+
+    Vector2d init_e = destination - start_pos;
+    float init_e_norm = init_e.norm();
+    Vector2d init_e_unit = init_e / init_e_norm;
+
+    // Reject near-zero desired speeds to avoid pedestrians that barely move,
+    // and reject negative speeds because they would reverse the intended
+    // walking direction.
+    float desired_v0;
+    do {
+      desired_v0 = velocity_gen(gen);
+    } while (desired_v0 <= 0.1);
+
+    float v_max = 1.3 * desired_v0;
+
+    P[i] = Pedestrian(BLACK, Vector2d(), start_pos, init_e_unit, v_max, desired_v0, destination);
+  }
+
+  return P;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+
+#if BENCHMARK
+  // Avoid compiler unused parameter warning
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+
+  // Each sample stores: time step, pedestrian id, x, y, e.x, e.y, color.
+  vector<tuple<int, int, float, float, float, float, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+
+    // Store the initial state before the first simulation step.
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+      samples.emplace_back(0, i, p.curr_pos.x, p.curr_pos.y, p.e.x, p.e.y, p.color);
+    }
+  }
+#endif
+
+  // # ADDs: n_iter * (
+  // #   n_p * 4 +
+  // #   n_p * (n_p - 1) * 14 +
+  // #   n_p * board.n_obstacles * 10 +
+  // #   n_p * 7
+  // # )
+
+  // # MULTs: n_iter * (
+  // #   n_p * 4
+  // #   n_p * (n_p - 1) * 24 +
+  // #   n_p * board.n_obstacles * 13 +
+  // #   n_p * 16
+  // # )
+
+  // # DIVs: n_iter * (
+  // #   n_p * (n_p - 1) +
+  // #   n_p * board.n_obstacles +
+  // #   n_p * 2
+  // # )
+
+  // # SQRTs: n_iter * (
+  // #   n_p * (n_p - 1) * 3 +
+  // #   n_p * board.n_obstacles +
+  // #   n_p * 3
+  // # )
+
+  // # EXPs: n_iter * (
+  // #   n_p * (n_p - 1) +
+  // #   n_p * board.n_obstacles
+  // # )
+
+  for (int t = 0; t < n_iter; t++) {
+    // Pass 1: Compute all forces from the current state.
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+
+      // # 2 ADDs, 4 MULTs
+      //
+      // $\vec{F}_\alpha(\vec{v}_\alpha, v^0_\alpha \vec{e}_\alpha) = \frac{1}{\tau_\alpha}(v^0_\alpha \vec{e}_\alpha - \vec{v}_\alpha)$
+      //
+      Vector2d f_alpha_dt = acceleration_term_dt(p);
+
+      // # (N-1) * (14 ADDs, 24 MULTs, 1 DIV, 3 SQRTs, 1 EXP)
+      //
+      // $\sum_{\beta} \vec{F}_{\alpha\beta}(\vec{e}_{\alpha},\vec{r}_{\alpha} - \vec{r}_{\beta})$
+      //
+      for (int j = 0; j < n_p; j++) {
+        if (i == j)
+          continue;
+
+        f_alpha_dt += pedestrian_repulsion_dt(p, pedestrians[j]);
+      }
+
+      // # board.n_obstacles * (10 ADDs, 13 MULTs, 1 DIV, 1 SQRT, 1 EXP)
+      //
+      // $\sum_{B} \vec{F}_{\alpha B}(\vec{e}_{\alpha}, \vec{r}_{\alpha} - \vec{r}_{B})$
+      //
+      for (int k = 0; k < board.n_obstacles; k++) {
+        f_alpha_dt += obstacle_repulsion_dt(p, board.obstacles[k]);
+      }
+
+      // # 2 ADDs
+      //
+      //
+      // $\frac{d\vec{w}_\alpha}{dt} = \vec{F}_{\alpha}(t)$
+      //
+      p.w += f_alpha_dt;
+    }
+
+    // Pass 2: Update pedestrian states after all forces have been computed.
+    // # (7 ADDs, 16 MULTs, 2 DIVs, 3 SQRTs) * n_p
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+
+      // # 1 ADD, 2 MULTs, 1 SQRT
+      float w_norm = p.w.norm();
+
+      float w_norm2 = w_norm > EPS ? w_norm : EPS; // Avoid division by zero.
+
+      // # 1 DIV
+      float correction_factor = std::min(1.0f, p.v_max / w_norm2);
+
+      // # 2 MULTs
+      p.curr_vel = p.w * correction_factor;
+
+      // # 2 ADDs, 2 MULTs
+      //
+      // $\vec{r}_i^{t+1} = \vec{r}_i^t + \vec{v}_i^t dt$
+      //
+      p.curr_pos = p.curr_pos + p.curr_vel * dt;
+
+      // When a pedestrian reaches the end of the corridor, teleport them back
+      // to the opposite entrance while keeping the same y-coordinate. This
+      // maintains a constant pedestrian flow and makes lane formation easier to
+      // observe.
+      if (p.color == WHITE && p.curr_pos.x > (BOARD_LENGTH - 1.5)) {
+        p.curr_pos.x = 1.0;
+        p.destination.y = p.curr_pos.y;
+        p.w = Vector2d(p.desired_v0, 0.0);
+        p.curr_vel = p.w;
+      }
+      if (p.color == BLACK && p.curr_pos.x < 1.5) {
+        p.curr_pos.x = BOARD_LENGTH - 1.0;
+        p.destination.y = p.curr_pos.y;
+        p.w = Vector2d(-p.desired_v0, 0.0);
+        p.curr_vel = p.w;
+      }
+
+      // Keep pedestrians inside the board boundaries.
+      p.curr_pos.y = clamp(p.curr_pos.y, WALL_EPS, BOARD_WIDTH_MINUS_WALL_EPS);
+
+      // # 2 ADDs
+      Vector2d e = p.destination - p.curr_pos;
+
+      // # 1 ADD, 2 MULTs, 1 DIV, 1 SQRT
+      float e_norm = e.norm();
+      float e_norm2 = e_norm > EPS ? e_norm : EPS; // Avoid division by zero.
+      float e_norm_inv = 1.0 / e_norm2;
+
+      // # 2 MULTs
+      p.e = e * e_norm_inv;
+
+      // # 1 ADD, 2 MULTs
+      float velocity_norm_sq = p.curr_vel.dot(p.curr_vel);
+
+      // # 1 SQRT
+      float velocity_norm = std::sqrt(velocity_norm_sq);
+
+      // # 1 MULT
+      p.s_b_sq = velocity_norm_sq * delta_t_sq;
+
+      // # 3 MULTs
+      p.s_b_e = p.e * (velocity_norm * delta_t);
+
+#if !BENCHMARK
+      // Store the current pedestrian state for visualization.
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, p.curr_pos.x, p.curr_pos.y, p.e.x, p.e.y, p.color);
+      }
+#endif
+    }
+  }
+
+#if !BENCHMARK
+  // Export recorded simulation states to CSV.
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
\ No newline at end of file
diff --git a/variants/variant_float_inlining_SoA_AVX2_polyexp_expfix.cpp b/variants/variant_float_inlining_SoA_AVX2_polyexp_expfix.cpp
index 2443933..572a2bc 100644
--- a/variants/variant_float_inlining_SoA_AVX2_polyexp_expfix.cpp
+++ b/variants/variant_float_inlining_SoA_AVX2_polyexp_expfix.cpp
@@ -180,6 +180,7 @@ static inline __m256 exp256_ps(__m256 x) {
   const __m256 ln2 = _mm256_set1_ps(0.6931471805599453f);
   __m256 r = _mm256_fnmadd_ps(nf, ln2, x); // # 8 MULTs, 8 ADDs
 
+  const __m256 c13 = _mm256_set1_ps(c13_scalar);
   const __m256 c12 = _mm256_set1_ps(c12_scalar);
   const __m256 c11 = _mm256_set1_ps(c11_scalar);
   const __m256 c10 = _mm256_set1_ps(c10_scalar);
@@ -194,7 +195,7 @@ static inline __m256 exp256_ps(__m256 x) {
   const __m256 c1  = _mm256_set1_ps(c1_scalar);
   const __m256 c0  = _mm256_set1_ps(c0_scalar);
 
-  __m256 p = c12;
+  __m256 p = c13;
 
   // #   13 * (8 MULTs, 8 ADDs)
   // # = 104 MULTs, 104 ADDs
diff --git a/variants/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix.cpp b/variants/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix.cpp
index 9c05350..0de7907 100644
--- a/variants/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix.cpp
+++ b/variants/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_expfix.cpp
@@ -197,6 +197,7 @@ static inline __m256 exp256_ps(__m256 x) {
   const __m256 ln2 = _mm256_set1_ps(0.6931471805599453f);
   __m256 r = _mm256_fnmadd_ps(nf, ln2, x); // # 8 MULTs, 8 ADDs
 
+  const __m256 c13 = _mm256_set1_ps(c13_scalar);
   const __m256 c12 = _mm256_set1_ps(c12_scalar);
   const __m256 c11 = _mm256_set1_ps(c11_scalar);
   const __m256 c10 = _mm256_set1_ps(c10_scalar);
@@ -211,7 +212,7 @@ static inline __m256 exp256_ps(__m256 x) {
   const __m256 c1  = _mm256_set1_ps(c1_scalar);
   const __m256 c0  = _mm256_set1_ps(c0_scalar);
 
-  __m256 p = c12;
+  __m256 p = c13;
 
   // #   13 * (8 MULTs, 8 ADDs)
   // # = 104 MULTs, 104 ADDs
diff --git a/variants/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix.cpp b/variants/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix.cpp
index c973f98..e6760fb 100644
--- a/variants/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix.cpp
+++ b/variants/variant_float_inlining_SoA_AVX2_polyexp_rsqrt_loopopt_expfix.cpp
@@ -197,6 +197,7 @@ static inline __m256 exp256_ps(__m256 x) {
   const __m256 ln2 = _mm256_set1_ps(0.6931471805599453f);
   __m256 r = _mm256_fnmadd_ps(nf, ln2, x); // # 8 MULTs, 8 ADDs
 
+  const __m256 c13 = _mm256_set1_ps(c13_scalar);
   const __m256 c12 = _mm256_set1_ps(c12_scalar);
   const __m256 c11 = _mm256_set1_ps(c11_scalar);
   const __m256 c10 = _mm256_set1_ps(c10_scalar);
@@ -211,7 +212,7 @@ static inline __m256 exp256_ps(__m256 x) {
   const __m256 c1 = _mm256_set1_ps(c1_scalar);
   const __m256 c0 = _mm256_set1_ps(c0_scalar);
 
-  __m256 p = c12;
+  __m256 p = c13;
 
   // #   13 * (8 MULTs, 8 ADDs)
   // # = 104 MULTs, 104 ADDs
diff --git a/variants/variant_float_inlining_SoA_AVX512_polyexp_expfix.cpp b/variants/variant_float_inlining_SoA_AVX512_polyexp_expfix.cpp
new file mode 100644
index 0000000..bcbeeea
--- /dev/null
+++ b/variants/variant_float_inlining_SoA_AVX512_polyexp_expfix.cpp
@@ -0,0 +1,540 @@
+// # Changes: Ported AVX2 to AVX512
+// # Predecessor: variant_float_inlining_SoA_AVX2_polyexp.cpp
+
+#include "../headers/board_reduced_float.hpp"
+#include "../headers/constants_float.hpp"
+#include "../headers/pedestrian_reduced_float.hpp"
+
+#include <immintrin.h>
+
+#include <fstream>
+#include <random>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0f
+#define BOARD_LENGTH 50.0f
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0f
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5f
+#define WALL_EPS 1e-3f
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999f
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+alignas(64) vector<float> pos_x, pos_y;
+alignas(64) vector<float> sb_sq;
+alignas(64) vector<float> sb_e_x, sb_e_y;
+alignas(64) vector<float> e_x, e_y;
+alignas(64) vector<float> curr_vel_x, curr_vel_y;
+alignas(64) vector<float> w_x, w_y;
+alignas(64) vector<float> desired_v0;
+alignas(64) vector<float> v_max;
+alignas(64) vector<float> destination_x, destination_y;
+alignas(64) vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<float> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<float> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<float> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<float> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  const int n_p_pad = (n_p + 15) & ~15;
+
+  pos_x.resize(n_p_pad);
+  pos_y.resize(n_p_pad);
+  sb_sq.resize(n_p_pad);
+  sb_e_x.resize(n_p_pad);
+  sb_e_y.resize(n_p_pad);
+  e_x.resize(n_p_pad);
+  e_y.resize(n_p_pad);
+  curr_vel_x.resize(n_p_pad);
+  curr_vel_y.resize(n_p_pad);
+  w_x.resize(n_p_pad);
+  w_y.resize(n_p_pad);
+  desired_v0.resize(n_p_pad);
+  v_max.resize(n_p_pad);
+  destination_x.resize(n_p_pad);
+  destination_y.resize(n_p_pad);
+  color.resize(n_p_pad);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    float x = x_gen_left(gen);
+    float y = y_gen(gen);
+
+    float dest_x = BOARD_LENGTH + 30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    float x = x_gen_right(gen);
+    float y = y_gen(gen);
+
+    float dest_x = -30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int j = n_p; j < n_p_pad; j++) {
+    pos_x[j] = 1e9f;
+    pos_y[j] = 1e9f;
+    sb_sq[j] = 0.0f;
+    sb_e_x[j] = 0.0f;
+    sb_e_y[j] = 0.0f;
+  }
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+// # 256 MULTs, 224 ADDs
+static inline __m512 _avx512_exp_ps(__m512 x) {
+
+  const __m512 log2e = _mm512_set1_ps(1.4426950408889634f);
+
+  // # 16 MULTs
+  __m512 nf = _mm512_roundscale_ps(_mm512_mul_ps(x, log2e), _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+
+  const __m512 ln2 = _mm512_set1_ps(0.6931471805599453f);
+
+  // # 16 MULTs, 16 ADDs
+  __m512 r = _mm512_fnmadd_ps(nf, ln2, x);
+
+  const __m512 c13 = _mm512_set1_ps(c13_scalar);
+  const __m512 c12 = _mm512_set1_ps(c12_scalar);
+  const __m512 c11 = _mm512_set1_ps(c11_scalar);
+  const __m512 c10 = _mm512_set1_ps(c10_scalar);
+  const __m512 c9 = _mm512_set1_ps(c9_scalar);
+  const __m512 c8 = _mm512_set1_ps(c8_scalar);
+  const __m512 c7 = _mm512_set1_ps(c7_scalar);
+  const __m512 c6 = _mm512_set1_ps(c6_scalar);
+  const __m512 c5 = _mm512_set1_ps(c5_scalar);
+  const __m512 c4 = _mm512_set1_ps(c4_scalar);
+  const __m512 c3 = _mm512_set1_ps(c3_scalar);
+  const __m512 c2 = _mm512_set1_ps(c2_scalar);
+  const __m512 c1 = _mm512_set1_ps(c1_scalar);
+  const __m512 c0 = _mm512_set1_ps(c0_scalar);
+
+  __m512 p = c13;
+
+  // #   13 * (16 MULTs, 16 ADDs)
+  // # = 208 MULTs, 208 ADDs
+  p = _mm512_fmadd_ps(p, r, c12);
+  p = _mm512_fmadd_ps(p, r, c11);
+  p = _mm512_fmadd_ps(p, r, c10);
+  p = _mm512_fmadd_ps(p, r, c9);
+  p = _mm512_fmadd_ps(p, r, c8);
+  p = _mm512_fmadd_ps(p, r, c7);
+  p = _mm512_fmadd_ps(p, r, c6);
+  p = _mm512_fmadd_ps(p, r, c5);
+  p = _mm512_fmadd_ps(p, r, c4);
+  p = _mm512_fmadd_ps(p, r, c3);
+  p = _mm512_fmadd_ps(p, r, c2);
+  p = _mm512_fmadd_ps(p, r, c1);
+  p = _mm512_fmadd_ps(p, r, c0);
+
+  __m512i ni = _mm512_cvtps_epi32(nf);
+  __m512i biased = _mm512_add_epi32(ni, _mm512_set1_epi32(127));
+  __m512i scaled = _mm512_slli_epi32(biased, 23);
+  __m512 pow2n = _mm512_castsi512_ps(scaled);
+
+  // # 16 MULTs
+  return _mm512_mul_ps(p, pow2n);
+}
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+  (void)pedestrians;
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+  vector<tuple<int, int, float, float, float, float, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+    for (int i = 0; i < n_p; i++) {
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  const int n_p_pad = (n_p + 15) & ~15;
+
+  alignas(64) const float lane_offset_arr[16] = {0.0f, 1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f};
+  const __m512 lane_off = _mm512_load_ps(lane_offset_arr);
+
+  const __m512 v_neg_v0_div_2s = _mm512_set1_ps(NEG_V0_AB_div_TWO_SIGMA);
+  const __m512 v_neg_2s_inv = _mm512_set1_ps(NEG_TWO_SIGMA_INV);
+  const __m512 v_cos_phi_sq = _mm512_set1_ps(cos_phi_sq);
+  const __m512 v_dt_neg = _mm512_set1_ps(dt_neg);
+  const __m512 v_c_neg_dt = _mm512_set1_ps(c_neg_times_dt);
+  const __m512 v_zero = _mm512_setzero_ps();
+  const __m512 v_one = _mm512_set1_ps(1.0f);
+  const __m512 v_eps = _mm512_set1_ps(EPS);
+  const __m512 v_eps_sq = _mm512_set1_ps(EPS_SQ);
+  const __m512 v_np = _mm512_set1_ps((float)n_p);
+  const __m512 v_exp_min = _mm512_set1_ps(-87.0f);
+
+  // # ADDs: n_iter * (
+  // #   n_p * 36 +
+  // #   n_p * n_p_pad * 29 +
+  // #   n_p * board.n_obstacles * 10 +
+  // #   n_p * 7
+  // # )
+
+  // # MULTs: n_iter * (
+  // #   n_p * 4 +
+  // #   n_p * n_p_pad * 40 +
+  // #   n_p * board.n_obstacles * 13 +
+  // #   n_p * 17
+  // # )
+
+  // # DIVs: n_iter * (
+  // #   n_p * n_p_pad +
+  // #   n_p * board.n_obstacles +
+  // #   n_p * 2
+  // # )
+
+  // # SQRTs: n_iter * (
+  // #   n_p * n_p_pad * 3 +
+  // #   n_p * board.n_obstacles +
+  // #   n_p * 3
+  // # )
+
+  // # EXPs: n_iter * n_p * board.n_obstacles
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+
+      // # 2 MULTs
+      float desired_velocity_x = e_x[i] * desired_v0[i];
+      float desired_velocity_y = e_y[i] * desired_v0[i];
+
+      // # 2 ADDs, 2 MULTs
+      float f_alpha_dt_x = (desired_velocity_x - curr_vel_x[i]) * dt_times_tau_alpha_inv;
+      float f_alpha_dt_y = (desired_velocity_y - curr_vel_y[i]) * dt_times_tau_alpha_inv;
+
+      const __m512 pix = _mm512_set1_ps(pos_x[i]);
+      const __m512 piy = _mm512_set1_ps(pos_y[i]);
+      const __m512 eix = _mm512_set1_ps(e_x[i]);
+      const __m512 eiy = _mm512_set1_ps(e_y[i]);
+      const __m512 v_i = _mm512_set1_ps((float)i);
+
+      __m512 acc_x = _mm512_setzero_ps();
+      __m512 acc_y = _mm512_setzero_ps();
+
+      for (int j = 0; j < n_p_pad; j += 16) {
+
+        // # 16 ADDs
+        __m512 j_idx = _mm512_add_ps(_mm512_set1_ps((float)j), lane_off);
+
+        __mmask16 m_neq_i = _mm512_cmp_ps_mask(j_idx, v_i, _CMP_NEQ_OQ);
+        __mmask16 m_in_range = _mm512_cmp_ps_mask(j_idx, v_np, _CMP_LT_OQ);
+        __mmask16 active = m_neq_i & m_in_range;
+
+        __m512 pjx = _mm512_loadu_ps(&pos_x[j]);
+        __m512 pjy = _mm512_loadu_ps(&pos_y[j]);
+        __m512 sbe_jx = _mm512_loadu_ps(&sb_e_x[j]);
+        __m512 sbe_jy = _mm512_loadu_ps(&sb_e_y[j]);
+        __m512 sb_jsq = _mm512_loadu_ps(&sb_sq[j]);
+
+        // # 32 ADDs
+        __m512 r_ab_x = _mm512_sub_ps(pix, pjx);
+        __m512 r_ab_y = _mm512_sub_ps(piy, pjy);
+
+        // # 32 MULTs, 16 ADDs
+        __m512 r_ab_n2 = _mm512_fmadd_ps(r_ab_x, r_ab_x, _mm512_mul_ps(r_ab_y, r_ab_y));
+
+        // # 16 SQRTs
+        __m512 r_ab_n = _mm512_sqrt_ps(r_ab_n2);
+
+        r_ab_n = _mm512_max_ps(r_ab_n, v_eps);
+
+        // # 32 ADDs
+        __m512 r_ab2_x = _mm512_sub_ps(r_ab_x, sbe_jx);
+        __m512 r_ab2_y = _mm512_sub_ps(r_ab_y, sbe_jy);
+
+        // # 32 MULTs, 16 ADDs
+        __m512 r_ab2_n2 = _mm512_fmadd_ps(r_ab2_x, r_ab2_x, _mm512_mul_ps(r_ab2_y, r_ab2_y));
+
+        // # 16 SQRTs
+        __m512 r_ab2_n = _mm512_sqrt_ps(r_ab2_n2);
+
+        r_ab2_n = _mm512_max_ps(r_ab2_n, v_eps);
+
+        // # 16 ADDs
+        __m512 r_sum = _mm512_add_ps(r_ab_n, r_ab2_n);
+
+        // # 16 MULTs, 16 ADDs
+        __m512 two_b_sq = _mm512_fmsub_ps(r_sum, r_sum, sb_jsq);
+
+        two_b_sq = _mm512_max_ps(two_b_sq, v_eps_sq);
+
+        // # 16 SQRTs
+        __m512 two_b = _mm512_sqrt_ps(two_b_sq);
+
+        // # 16 MULTs
+        __m512 e_arg = _mm512_mul_ps(two_b, v_neg_2s_inv);
+
+        e_arg = _mm512_max_ps(e_arg, v_exp_min);
+
+        // # 256 MULTs, 224 ADDs
+        __m512 e_val = _avx512_exp_ps(e_arg);
+
+        // # 64 MULTs
+        __m512 numer = _mm512_mul_ps(_mm512_mul_ps(r_sum, v_neg_v0_div_2s), e_val);
+        __m512 denom = _mm512_mul_ps(_mm512_mul_ps(two_b, r_ab_n), r_ab2_n);
+
+        __m512 denom_safe = _mm512_mask_blend_ps(active, v_one, denom);
+
+        // # 16 DIVs
+        __m512 smc = _mm512_div_ps(numer, denom_safe);
+
+        // # 32 MULTs
+        __m512 sm_r_ab = _mm512_mul_ps(smc, r_ab2_n);
+        __m512 sm_r_ab2 = _mm512_mul_ps(smc, r_ab_n);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 f_x = _mm512_fmadd_ps(r_ab2_x, sm_r_ab2, _mm512_mul_ps(r_ab_x, sm_r_ab));
+        __m512 f_y = _mm512_fmadd_ps(r_ab2_y, sm_r_ab2, _mm512_mul_ps(r_ab_y, sm_r_ab));
+
+        // # 32 MULTs, 16 ADDs
+        __m512 f_norm_sq = _mm512_fmadd_ps(f_x, f_x, _mm512_mul_ps(f_y, f_y));
+
+        // # 16 MULTs
+        __m512 rhs = _mm512_mul_ps(f_norm_sq, v_cos_phi_sq);
+
+        // # 32 MULTs, 16 ADDs
+        __m512 e_dot_f = _mm512_fmadd_ps(eix, f_x, _mm512_mul_ps(eiy, f_y));
+
+        // # 16 MULTs
+        __m512 e_dot_f_sq = _mm512_mul_ps(e_dot_f, e_dot_f);
+
+        __mmask16 c1 = _mm512_cmp_ps_mask(e_dot_f, v_zero, _CMP_LT_OQ);
+        __mmask16 c2 = _mm512_cmp_ps_mask(e_dot_f_sq, rhs, _CMP_GT_OQ);
+        __mmask16 cond = c1 & c2;
+
+        __m512 w_val = _mm512_mask_blend_ps(cond, v_dt_neg, v_c_neg_dt);
+
+        __m512 f_x_safe = _mm512_maskz_mov_ps(active, f_x);
+        __m512 f_y_safe = _mm512_maskz_mov_ps(active, f_y);
+
+        // # 32 MULTs, 32 ADDs
+        acc_x = _mm512_fmadd_ps(f_x_safe, w_val, acc_x);
+        acc_y = _mm512_fmadd_ps(f_y_safe, w_val, acc_y);
+      }
+
+      // # 32 ADDs
+      f_alpha_dt_x += _mm512_reduce_add_ps(acc_x);
+      f_alpha_dt_y += _mm512_reduce_add_ps(acc_y);
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
+        float v1_x = pos_x[i] - board.obstacles[k].start_x;
+        float v1_y = pos_y[i] - board.obstacles[k].start_y;
+        float v2_x = board.obstacles[k].obstacle_v_x;
+        float v2_y = board.obstacles[k].obstacle_v_y;
+
+        // # 3 MULTs, 1 ADD
+        float projection = (v1_x * v2_x + v1_y * v2_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0f)
+          projection = 0.0f;
+        else if (projection > 1.0f)
+          projection = 1.0f;
+
+        // # 4 ADDs, 2 MULTs
+        float r_aB_x = pos_x[i] - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        float r_aB_y = pos_y[i] - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        float r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+        r_aB_norm = r_aB_norm > EPS ? r_aB_norm : EPS;
+
+        // # 4 MULTs, 1 EXP, 1 DIV
+        float scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+        float f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        float f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        float w_val = dt_neg;
+        float e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -e_y[i] : e_y[i]);
+
+        if (e_dot_f < cos_phi)
+          w_val = c_neg_times_dt;
+
+        // # 2 MULTS
+        float f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        float f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        // # 2 ADDs
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p; i++) {
+
+      // # 1 SQRT, 2 MULTs, 1 ADD
+      float w_norm = sqrt(w_x[i] * w_x[i] + w_y[i] * w_y[i]);
+
+      float w_norm2 = w_norm > EPS ? w_norm : EPS;
+
+      // # 1 DIV
+      float correction_factor = min(1.0f, v_max[i] / w_norm2);
+
+      // # 2 MULTs
+      curr_vel_x[i] = w_x[i] * correction_factor;
+      curr_vel_y[i] = w_y[i] * correction_factor;
+
+      // # 2 MULTs, 2 ADDs
+      pos_x[i] = pos_x[i] + curr_vel_x[i] * dt;
+      pos_y[i] = pos_y[i] + curr_vel_y[i] * dt;
+
+      if (color[i] == 0 && pos_x[i] > BOARD_LENGTH_MINUS_ONE_POINT_FIVE) {
+        pos_x[i] = 1.0f;
+        destination_y[i] = pos_y[i];
+        w_x[i] = desired_v0[i];
+        w_y[i] = 0.0f;
+        curr_vel_x[i] = w_x[i];
+        curr_vel_y[i] = w_y[i];
+      }
+      if (color[i] == 1 && pos_x[i] < 1.5f) {
+        pos_x[i] = BOARD_LENGTH_MINUS_ONE;
+        destination_y[i] = pos_y[i];
+        w_x[i] = -desired_v0[i];
+        w_y[i] = 0.0f;
+        curr_vel_x[i] = w_x[i];
+        curr_vel_y[i] = w_y[i];
+      }
+
+      pos_y[i] = clamp(pos_y[i], WALL_EPS, BOARD_WIDTH_MINUS_WALL_EPS);
+
+      // # 2 ADDs
+      float ex = destination_x[i] - pos_x[i];
+      float ey = destination_y[i] - pos_y[i];
+
+      // # 1 SQRT, 2 MULTs, 1 ADD
+      float e_norm = sqrt(ex * ex + ey * ey);
+
+      float e_norm2 = e_norm > EPS ? e_norm : EPS;
+
+      // # 1 DIV
+      float e_norm_inv = 1.0f / e_norm2;
+
+      // # 2 MULTs
+      e_x[i] = ex * e_norm_inv;
+      e_y[i] = ey * e_norm_inv;
+
+      // # 2 MULTs, 1 ADD
+      float velocity_norm_sq = curr_vel_x[i] * curr_vel_x[i] + curr_vel_y[i] * curr_vel_y[i];
+
+      // # 1 SQRT
+      float velocity_norm = sqrt(velocity_norm_sq);
+
+      // # 5 MULTs
+      sb_sq[i] = velocity_norm_sq * delta_t_sq;
+      sb_e_x[i] = e_x[i] * (velocity_norm * delta_t);
+      sb_e_y[i] = e_y[i] * (velocity_norm * delta_t);
+
+#if !BENCHMARK
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+#endif
+    }
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp b/variants/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp
new file mode 100644
index 0000000..f0ab633
--- /dev/null
+++ b/variants/variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp
@@ -0,0 +1,557 @@
+// # Changes: Added more correction rounds to _avx512_exp_ps to do 30 flops per scalar
+// # Predecessor: variant_float_inlining_SoA_AVX512_polyexp_fastinvsqrt.cpp
+
+#include "../headers/board_reduced_float.hpp"
+#include "../headers/constants_float.hpp"
+#include "../headers/pedestrian_reduced_float.hpp"
+
+#include <immintrin.h>
+
+#include <fstream>
+#include <random>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0f
+#define BOARD_LENGTH 50.0f
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0f
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5f
+#define WALL_EPS 1e-3f
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999f
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+alignas(64) vector<float> pos_x, pos_y;
+alignas(64) vector<float> sb_sq;
+alignas(64) vector<float> sb_e_x, sb_e_y;
+alignas(64) vector<float> e_x, e_y;
+alignas(64) vector<float> curr_vel_x, curr_vel_y;
+alignas(64) vector<float> w_x, w_y;
+alignas(64) vector<float> desired_v0;
+alignas(64) vector<float> v_max;
+alignas(64) vector<float> destination_x, destination_y;
+alignas(64) vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<float> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<float> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<float> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<float> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  const int n_p_pad = (n_p + 15) & ~15;
+
+  pos_x.resize(n_p_pad);
+  pos_y.resize(n_p_pad);
+  sb_sq.resize(n_p_pad);
+  sb_e_x.resize(n_p_pad);
+  sb_e_y.resize(n_p_pad);
+  e_x.resize(n_p_pad);
+  e_y.resize(n_p_pad);
+  curr_vel_x.resize(n_p_pad);
+  curr_vel_y.resize(n_p_pad);
+  w_x.resize(n_p_pad);
+  w_y.resize(n_p_pad);
+  desired_v0.resize(n_p_pad);
+  v_max.resize(n_p_pad);
+  destination_x.resize(n_p_pad);
+  destination_y.resize(n_p_pad);
+  color.resize(n_p_pad);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    float x = x_gen_left(gen);
+    float y = y_gen(gen);
+
+    float dest_x = BOARD_LENGTH + 30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    float x = x_gen_right(gen);
+    float y = y_gen(gen);
+
+    float dest_x = -30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int j = n_p; j < n_p_pad; j++) {
+    pos_x[j] = 1e9f;
+    pos_y[j] = 1e9f;
+    sb_sq[j] = 0.0f;
+    sb_e_x[j] = 0.0f;
+    sb_e_y[j] = 0.0f;
+  }
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+// # 256 MULTs, 224 ADDs
+static inline __m512 _avx512_exp_ps(__m512 x) {
+
+  const __m512 log2e = _mm512_set1_ps(1.4426950408889634f);
+
+  // # 16 MULTs
+  __m512 nf = _mm512_roundscale_ps(_mm512_mul_ps(x, log2e), _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+
+  const __m512 ln2 = _mm512_set1_ps(0.6931471805599453f);
+
+  // # 16 MULTs, 16 ADDs
+  __m512 r = _mm512_fnmadd_ps(nf, ln2, x);
+
+  const __m512 c13 = _mm512_set1_ps(c13_scalar);
+  const __m512 c12 = _mm512_set1_ps(c12_scalar);
+  const __m512 c11 = _mm512_set1_ps(c11_scalar);
+  const __m512 c10 = _mm512_set1_ps(c10_scalar);
+  const __m512 c9 = _mm512_set1_ps(c9_scalar);
+  const __m512 c8 = _mm512_set1_ps(c8_scalar);
+  const __m512 c7 = _mm512_set1_ps(c7_scalar);
+  const __m512 c6 = _mm512_set1_ps(c6_scalar);
+  const __m512 c5 = _mm512_set1_ps(c5_scalar);
+  const __m512 c4 = _mm512_set1_ps(c4_scalar);
+  const __m512 c3 = _mm512_set1_ps(c3_scalar);
+  const __m512 c2 = _mm512_set1_ps(c2_scalar);
+  const __m512 c1 = _mm512_set1_ps(c1_scalar);
+  const __m512 c0 = _mm512_set1_ps(c0_scalar);
+
+  __m512 p = c13;
+
+  // #   13 * (16 MULTs, 16 ADDs)
+  // # = 208 MULTs, 208 ADDs
+  p = _mm512_fmadd_ps(p, r, c12);
+  p = _mm512_fmadd_ps(p, r, c11);
+  p = _mm512_fmadd_ps(p, r, c10);
+  p = _mm512_fmadd_ps(p, r, c9);
+  p = _mm512_fmadd_ps(p, r, c8);
+  p = _mm512_fmadd_ps(p, r, c7);
+  p = _mm512_fmadd_ps(p, r, c6);
+  p = _mm512_fmadd_ps(p, r, c5);
+  p = _mm512_fmadd_ps(p, r, c4);
+  p = _mm512_fmadd_ps(p, r, c3);
+  p = _mm512_fmadd_ps(p, r, c2);
+  p = _mm512_fmadd_ps(p, r, c1);
+  p = _mm512_fmadd_ps(p, r, c0);
+
+  __m512i ni = _mm512_cvtps_epi32(nf);
+  __m512i biased = _mm512_add_epi32(ni, _mm512_set1_epi32(127));
+  __m512i scaled = _mm512_slli_epi32(biased, 23);
+  __m512 pow2n = _mm512_castsi512_ps(scaled);
+
+  // # 16 MULTs
+  return _mm512_mul_ps(p, pow2n);
+}
+
+// # 7 MULTs, 2 ADDs
+static inline float fast_inv_sqrt(float number) {
+  float y = number;
+
+  uint32_t i = bit_cast<uint32_t>(y);
+
+  i = 0x5f3759df - (i >> 1);
+
+  y = bit_cast<float>(i);
+
+  // # 1 MULT
+  float x2 = number * 0.5;
+
+  // # 6 MULTs, 2 ADDs
+  y = y * (1.5 - (x2 * y * y));
+  y = y * (1.5 - (x2 * y * y));
+
+  return y;
+}
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+  (void)pedestrians;
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+  vector<tuple<int, int, float, float, float, float, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+    for (int i = 0; i < n_p; i++) {
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  const int n_p_pad = (n_p + 15) & ~15;
+
+  alignas(64) const float lane_offset_arr[16] = {0.0f, 1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f};
+  const __m512 lane_off = _mm512_load_ps(lane_offset_arr);
+
+  const __m512 v_neg_v0_div_2s = _mm512_set1_ps(NEG_V0_AB_div_TWO_SIGMA);
+  const __m512 v_neg_2s_inv = _mm512_set1_ps(NEG_TWO_SIGMA_INV);
+  const __m512 v_cos_phi_sq = _mm512_set1_ps(cos_phi_sq);
+  const __m512 v_dt_neg = _mm512_set1_ps(dt_neg);
+  const __m512 v_c_neg_dt = _mm512_set1_ps(c_neg_times_dt);
+  const __m512 v_zero = _mm512_setzero_ps();
+  const __m512 v_one = _mm512_set1_ps(1.0f);
+  const __m512 v_eps = _mm512_set1_ps(EPS);
+  const __m512 v_eps_sq = _mm512_set1_ps(EPS_SQ);
+  const __m512 v_np = _mm512_set1_ps((float)n_p);
+  const __m512 v_exp_min = _mm512_set1_ps(-87.0f);
+
+  // # ADDs: n_iter * (
+  // #   n_p * 36 +
+  // #   n_p * n_p_pad * 29 +
+  // #   n_p * board.n_obstacles * 10 +
+  // #   n_p * 11
+  // # )
+
+  // # MULTs: n_iter * (
+  // #   n_p * 4 +
+  // #   n_p * n_p_pad * 40 +
+  // #   n_p * board.n_obstacles * 13 +
+  // #   n_p * 30
+  // # )
+
+  // # DIVs: n_iter * (
+  // #   n_p * n_p_pad +
+  // #   n_p * board.n_obstacles
+  // # )
+
+  // # SQRTs: n_iter * (
+  // #   n_p * n_p_pad * 3 +
+  // #   n_p * board.n_obstacles +
+  // #   n_p
+  // # )
+
+  // # EXPs: n_iter * n_p * board.n_obstacles
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+
+      // # 2 MULTs
+      float desired_velocity_x = e_x[i] * desired_v0[i];
+      float desired_velocity_y = e_y[i] * desired_v0[i];
+
+      // # 2 ADDs, 2 MULTs
+      float f_alpha_dt_x = (desired_velocity_x - curr_vel_x[i]) * dt_times_tau_alpha_inv;
+      float f_alpha_dt_y = (desired_velocity_y - curr_vel_y[i]) * dt_times_tau_alpha_inv;
+
+      const __m512 pix = _mm512_set1_ps(pos_x[i]);
+      const __m512 piy = _mm512_set1_ps(pos_y[i]);
+      const __m512 eix = _mm512_set1_ps(e_x[i]);
+      const __m512 eiy = _mm512_set1_ps(e_y[i]);
+      const __m512 v_i = _mm512_set1_ps((float)i);
+
+      __m512 acc_x = _mm512_setzero_ps();
+      __m512 acc_y = _mm512_setzero_ps();
+
+      for (int j = 0; j < n_p_pad; j += 16) {
+
+        // # 16 ADDs
+        __m512 j_idx = _mm512_add_ps(_mm512_set1_ps((float)j), lane_off);
+
+        __mmask16 m_neq_i = _mm512_cmp_ps_mask(j_idx, v_i, _CMP_NEQ_OQ);
+        __mmask16 m_in_range = _mm512_cmp_ps_mask(j_idx, v_np, _CMP_LT_OQ);
+        __mmask16 active = m_neq_i & m_in_range;
+
+        __m512 pjx = _mm512_loadu_ps(&pos_x[j]);
+        __m512 pjy = _mm512_loadu_ps(&pos_y[j]);
+        __m512 sbe_jx = _mm512_loadu_ps(&sb_e_x[j]);
+        __m512 sbe_jy = _mm512_loadu_ps(&sb_e_y[j]);
+        __m512 sb_jsq = _mm512_loadu_ps(&sb_sq[j]);
+
+        // # 32 ADDs
+        __m512 r_ab_x = _mm512_sub_ps(pix, pjx);
+        __m512 r_ab_y = _mm512_sub_ps(piy, pjy);
+
+        // # 32 MULTs, 16 ADDs
+        __m512 r_ab_n2 = _mm512_fmadd_ps(r_ab_x, r_ab_x, _mm512_mul_ps(r_ab_y, r_ab_y));
+
+        // # 16 SQRTs
+        __m512 r_ab_n = _mm512_sqrt_ps(r_ab_n2);
+
+        r_ab_n = _mm512_max_ps(r_ab_n, v_eps);
+
+        // # 32 ADDs
+        __m512 r_ab2_x = _mm512_sub_ps(r_ab_x, sbe_jx);
+        __m512 r_ab2_y = _mm512_sub_ps(r_ab_y, sbe_jy);
+
+        // # 32 MULTs, 16 ADDs
+        __m512 r_ab2_n2 = _mm512_fmadd_ps(r_ab2_x, r_ab2_x, _mm512_mul_ps(r_ab2_y, r_ab2_y));
+
+        // # 16 SQRTs
+        __m512 r_ab2_n = _mm512_sqrt_ps(r_ab2_n2);
+
+        r_ab2_n = _mm512_max_ps(r_ab2_n, v_eps);
+
+        // # 16 ADDs
+        __m512 r_sum = _mm512_add_ps(r_ab_n, r_ab2_n);
+
+        // # 16 MULTs, 16 ADDs
+        __m512 two_b_sq = _mm512_fmsub_ps(r_sum, r_sum, sb_jsq);
+
+        two_b_sq = _mm512_max_ps(two_b_sq, v_eps_sq);
+
+        // # 16 SQRTs
+        __m512 two_b = _mm512_sqrt_ps(two_b_sq);
+
+        // # 16 MULTs
+        __m512 e_arg = _mm512_mul_ps(two_b, v_neg_2s_inv);
+
+        e_arg = _mm512_max_ps(e_arg, v_exp_min);
+
+        // # 256 MULTs, 224 ADDs
+        __m512 e_val = _avx512_exp_ps(e_arg);
+
+        // # 64 MULTs
+        __m512 numer = _mm512_mul_ps(_mm512_mul_ps(r_sum, v_neg_v0_div_2s), e_val);
+        __m512 denom = _mm512_mul_ps(_mm512_mul_ps(two_b, r_ab_n), r_ab2_n);
+
+        __m512 denom_safe = _mm512_mask_blend_ps(active, v_one, denom);
+
+        // # 16 DIVs
+        __m512 smc = _mm512_div_ps(numer, denom_safe);
+
+        // # 32 MULTs
+        __m512 sm_r_ab = _mm512_mul_ps(smc, r_ab2_n);
+        __m512 sm_r_ab2 = _mm512_mul_ps(smc, r_ab_n);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 f_x = _mm512_fmadd_ps(r_ab2_x, sm_r_ab2, _mm512_mul_ps(r_ab_x, sm_r_ab));
+        __m512 f_y = _mm512_fmadd_ps(r_ab2_y, sm_r_ab2, _mm512_mul_ps(r_ab_y, sm_r_ab));
+
+        // # 32 MULTs, 16 ADDs
+        __m512 f_norm_sq = _mm512_fmadd_ps(f_x, f_x, _mm512_mul_ps(f_y, f_y));
+
+        // # 16 MULTs
+        __m512 rhs = _mm512_mul_ps(f_norm_sq, v_cos_phi_sq);
+
+        // # 32 MULTs, 16 ADDs
+        __m512 e_dot_f = _mm512_fmadd_ps(eix, f_x, _mm512_mul_ps(eiy, f_y));
+
+        // # 16 MULTs
+        __m512 e_dot_f_sq = _mm512_mul_ps(e_dot_f, e_dot_f);
+
+        __mmask16 c1 = _mm512_cmp_ps_mask(e_dot_f, v_zero, _CMP_LT_OQ);
+        __mmask16 c2 = _mm512_cmp_ps_mask(e_dot_f_sq, rhs, _CMP_GT_OQ);
+        __mmask16 cond = c1 & c2;
+
+        __m512 w_val = _mm512_mask_blend_ps(cond, v_dt_neg, v_c_neg_dt);
+
+        __m512 f_x_safe = _mm512_maskz_mov_ps(active, f_x);
+        __m512 f_y_safe = _mm512_maskz_mov_ps(active, f_y);
+
+        // # 32 MULTs, 32 ADDs
+        acc_x = _mm512_fmadd_ps(f_x_safe, w_val, acc_x);
+        acc_y = _mm512_fmadd_ps(f_y_safe, w_val, acc_y);
+      }
+
+      // # 32 ADDs
+      f_alpha_dt_x += _mm512_reduce_add_ps(acc_x);
+      f_alpha_dt_y += _mm512_reduce_add_ps(acc_y);
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
+        float v1_x = pos_x[i] - board.obstacles[k].start_x;
+        float v1_y = pos_y[i] - board.obstacles[k].start_y;
+        float v2_x = board.obstacles[k].obstacle_v_x;
+        float v2_y = board.obstacles[k].obstacle_v_y;
+
+        // # 3 MULTs, 1 ADD
+        float projection = (v1_x * v2_x + v1_y * v2_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0f)
+          projection = 0.0f;
+        else if (projection > 1.0f)
+          projection = 1.0f;
+
+        // # 4 ADDs, 2 MULTs
+        float r_aB_x = pos_x[i] - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        float r_aB_y = pos_y[i] - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        float r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+
+        // # 2 MULTs, 1 EXP, 1 DIV
+        float scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+
+        // # 2 MULTs
+        float f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        float f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        float w_val = dt_neg;
+        float e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -e_y[i] : e_y[i]);
+
+        if (e_dot_f < cos_phi)
+          w_val = c_neg_times_dt;
+
+        // # 2 MULTs
+        float f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        float f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        // # 2 ADDs
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      // # 2 ADDs
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p; i++) {
+      // # 9 MULTs, 3 ADDs
+      float w_norm_inv = fast_inv_sqrt(w_x[i] * w_x[i] + w_y[i] * w_y[i]);
+
+      // # 1 MULT
+      float correction_factor = min(1.0f, v_max[i] * w_norm_inv);
+
+      // # 2 MULTs
+      curr_vel_x[i] = w_x[i] * correction_factor;
+      curr_vel_y[i] = w_y[i] * correction_factor;
+
+      // # 2 MULTs, 2 ADDs
+      pos_x[i] = pos_x[i] + curr_vel_x[i] * dt;
+      pos_y[i] = pos_y[i] + curr_vel_y[i] * dt;
+
+      if (color[i] == 0 && pos_x[i] > (BOARD_LENGTH_MINUS_ONE_POINT_FIVE)) {
+        pos_x[i] = 1.0f;
+        destination_y[i] = pos_y[i];
+        w_x[i] = desired_v0[i];
+        w_y[i] = 0.0f;
+        curr_vel_x[i] = w_x[i];
+        curr_vel_y[i] = w_y[i];
+      }
+      if (color[i] == 1 && pos_x[i] < 1.5f) {
+        pos_x[i] = BOARD_LENGTH_MINUS_ONE;
+        destination_y[i] = pos_y[i];
+        w_x[i] = -desired_v0[i];
+        w_y[i] = 0.0f;
+        curr_vel_x[i] = w_x[i];
+        curr_vel_y[i] = w_y[i];
+      }
+
+      pos_y[i] = clamp(pos_y[i], WALL_EPS, BOARD_WIDTH_MINUS_WALL_EPS);
+
+      // # 2 ADDs
+      float ex = destination_x[i] - pos_x[i];
+      float ey = destination_y[i] - pos_y[i];
+
+      // # 9 MULTs, 3 ADDs
+      float e_norm_inv = fast_inv_sqrt(ex * ex + ey * ey);
+
+      // # 2 MULTs
+      e_x[i] = ex * e_norm_inv;
+      e_y[i] = ey * e_norm_inv;
+
+      // # 2 MULTs, 1 ADD
+      double velocity_norm_sq = curr_vel_x[i] * curr_vel_x[i] + curr_vel_y[i] * curr_vel_y[i];
+
+      // # 1 MULT
+      double sb_sq_i = velocity_norm_sq * delta_t_sq;
+
+      // # 1 SQRT
+      double velocity_norm_mul_delta_t = sqrt(sb_sq_i);
+
+      sb_sq[i] = sb_sq_i;
+
+      // # 2 MULTs
+      sb_e_x[i] = e_x[i] * velocity_norm_mul_delta_t;
+      sb_e_y[i] = e_y[i] * velocity_norm_mul_delta_t;
+
+#if !BENCHMARK
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+#endif
+    }
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix.cpp b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix.cpp
new file mode 100644
index 0000000..720810a
--- /dev/null
+++ b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_expfix.cpp
@@ -0,0 +1,577 @@
+// # Changes: Added more correction rounds to _avx512_exp_ps to do 30 flops per scalar
+// # Predecessor: variant_float_inlining_SoA_AVX512_polyexp_moreavx.cpp
+
+#include "../headers/board_reduced_float.hpp"
+#include "../headers/constants_float.hpp"
+#include "../headers/pedestrian_reduced_float.hpp"
+
+#include <immintrin.h>
+
+#include <fstream>
+#include <random>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0f
+#define BOARD_LENGTH 50.0f
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0f
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5f
+#define WALL_EPS 1e-3f
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999f
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+alignas(64) vector<float> pos_x, pos_y;
+alignas(64) vector<float> sb_sq;
+alignas(64) vector<float> sb_e_x, sb_e_y;
+alignas(64) vector<float> e_x, e_y;
+alignas(64) vector<float> curr_vel_x, curr_vel_y;
+alignas(64) vector<float> w_x, w_y;
+alignas(64) vector<float> desired_v0;
+alignas(64) vector<float> v_max;
+alignas(64) vector<float> destination_x, destination_y;
+alignas(64) vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<float> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<float> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<float> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<float> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  const int n_p_pad = (n_p + 15) & ~15;
+
+  pos_x.resize(n_p_pad);
+  pos_y.resize(n_p_pad);
+  sb_sq.resize(n_p_pad);
+  sb_e_x.resize(n_p_pad);
+  sb_e_y.resize(n_p_pad);
+  e_x.resize(n_p_pad);
+  e_y.resize(n_p_pad);
+  curr_vel_x.resize(n_p_pad);
+  curr_vel_y.resize(n_p_pad);
+  w_x.resize(n_p_pad);
+  w_y.resize(n_p_pad);
+  desired_v0.resize(n_p_pad);
+  v_max.resize(n_p_pad);
+  destination_x.resize(n_p_pad);
+  destination_y.resize(n_p_pad);
+  color.resize(n_p_pad);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    float x = x_gen_left(gen);
+    float y = y_gen(gen);
+
+    float dest_x = BOARD_LENGTH + 30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    float x = x_gen_right(gen);
+    float y = y_gen(gen);
+
+    float dest_x = -30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int j = n_p; j < n_p_pad; j++) {
+    pos_x[j] = 1e9f;
+    pos_y[j] = 1e9f;
+    sb_sq[j] = 0.0f;
+    sb_e_x[j] = 0.0f;
+    sb_e_y[j] = 0.0f;
+  }
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+// # 256 MULTs, 224 ADDs
+static inline __m512 _avx512_exp_ps(__m512 x) {
+
+  const __m512 log2e = _mm512_set1_ps(1.4426950408889634f);
+
+  // # 16 MULTs
+  __m512 nf = _mm512_roundscale_ps(_mm512_mul_ps(x, log2e), _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+
+  const __m512 ln2 = _mm512_set1_ps(0.6931471805599453f);
+
+  // # 16 MULTs, 16 ADDs
+  __m512 r = _mm512_fnmadd_ps(nf, ln2, x);
+
+  const __m512 c13 = _mm512_set1_ps(c13_scalar);
+  const __m512 c12 = _mm512_set1_ps(c12_scalar);
+  const __m512 c11 = _mm512_set1_ps(c11_scalar);
+  const __m512 c10 = _mm512_set1_ps(c10_scalar);
+  const __m512 c9 = _mm512_set1_ps(c9_scalar);
+  const __m512 c8 = _mm512_set1_ps(c8_scalar);
+  const __m512 c7 = _mm512_set1_ps(c7_scalar);
+  const __m512 c6 = _mm512_set1_ps(c6_scalar);
+  const __m512 c5 = _mm512_set1_ps(c5_scalar);
+  const __m512 c4 = _mm512_set1_ps(c4_scalar);
+  const __m512 c3 = _mm512_set1_ps(c3_scalar);
+  const __m512 c2 = _mm512_set1_ps(c2_scalar);
+  const __m512 c1 = _mm512_set1_ps(c1_scalar);
+  const __m512 c0 = _mm512_set1_ps(c0_scalar);
+
+  __m512 p = c13;
+
+  // #   13 * (16 MULTs, 16 ADDs)
+  // # = 208 MULTs, 208 ADDs
+  p = _mm512_fmadd_ps(p, r, c12);
+  p = _mm512_fmadd_ps(p, r, c11);
+  p = _mm512_fmadd_ps(p, r, c10);
+  p = _mm512_fmadd_ps(p, r, c9);
+  p = _mm512_fmadd_ps(p, r, c8);
+  p = _mm512_fmadd_ps(p, r, c7);
+  p = _mm512_fmadd_ps(p, r, c6);
+  p = _mm512_fmadd_ps(p, r, c5);
+  p = _mm512_fmadd_ps(p, r, c4);
+  p = _mm512_fmadd_ps(p, r, c3);
+  p = _mm512_fmadd_ps(p, r, c2);
+  p = _mm512_fmadd_ps(p, r, c1);
+  p = _mm512_fmadd_ps(p, r, c0);
+
+  __m512i ni = _mm512_cvtps_epi32(nf);
+  __m512i biased = _mm512_add_epi32(ni, _mm512_set1_epi32(127));
+  __m512i scaled = _mm512_slli_epi32(biased, 23);
+  __m512 pow2n = _mm512_castsi512_ps(scaled);
+
+  // # 16 MULTs
+  return _mm512_mul_ps(p, pow2n);
+}
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+  (void)pedestrians;
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+  vector<tuple<int, int, float, float, float, float, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+    for (int i = 0; i < n_p; i++) {
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  const int n_p_pad = (n_p + 15) & ~15;
+
+  alignas(64) const float lane_offset_arr[16] = {0.0f, 1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f};
+  const __m512 lane_off = _mm512_load_ps(lane_offset_arr);
+
+  const __m512 v_neg_v0_div_2s = _mm512_set1_ps(NEG_V0_AB_div_TWO_SIGMA);
+  const __m512 v_neg_2s_inv = _mm512_set1_ps(NEG_TWO_SIGMA_INV);
+  const __m512 v_cos_phi_sq = _mm512_set1_ps(cos_phi_sq);
+  const __m512 v_dt_neg = _mm512_set1_ps(dt_neg);
+  const __m512 v_c_neg_dt = _mm512_set1_ps(c_neg_times_dt);
+  const __m512 v_zero = _mm512_setzero_ps();
+  const __m512 v_one = _mm512_set1_ps(1.0f);
+  const __m512 v_eps = _mm512_set1_ps(EPS);
+  const __m512 v_eps_sq = _mm512_set1_ps(EPS_SQ);
+  const __m512 v_np = _mm512_set1_ps((float)n_p);
+  const __m512 v_exp_min = _mm512_set1_ps(-87.0f);
+
+  // # ADDs: n_iter * (
+  // #   n_p * 36 +
+  // #   n_p * n_p_pad * 29 +
+  // #   n_p * board.n_obstacles * 10 +
+  // #   n_p_pad * 7
+  // # )
+
+  // # MULTs: n_iter * (
+  // #   n_p * 4 +
+  // #   n_p * n_p_pad * 40 +
+  // #   n_p * board.n_obstacles * 13 +
+  // #   n_p_pad * 15
+  // # )
+
+  // # DIVs: n_iter * (
+  // #   n_p * n_p_pad +
+  // #   n_p * board.n_obstacles +
+  // #   n_p_pad * 2
+  // # )
+
+  // # SQRTs: n_iter * (
+  // #   n_p * n_p_pad * 3 +
+  // #   n_p * board.n_obstacles +
+  // #   n_p_pad * 3
+  // # )
+
+  // # EXPs: n_iter * n_p * board.n_obstacles
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+
+      // # 2 MULTs
+      float desired_velocity_x = e_x[i] * desired_v0[i];
+      float desired_velocity_y = e_y[i] * desired_v0[i];
+
+      // # 2 ADDs, 2 MULTs
+      float f_alpha_dt_x = (desired_velocity_x - curr_vel_x[i]) * dt_times_tau_alpha_inv;
+      float f_alpha_dt_y = (desired_velocity_y - curr_vel_y[i]) * dt_times_tau_alpha_inv;
+
+      const __m512 pix = _mm512_set1_ps(pos_x[i]);
+      const __m512 piy = _mm512_set1_ps(pos_y[i]);
+      const __m512 eix = _mm512_set1_ps(e_x[i]);
+      const __m512 eiy = _mm512_set1_ps(e_y[i]);
+      const __m512 v_i = _mm512_set1_ps((float)i);
+
+      __m512 acc_x = _mm512_setzero_ps();
+      __m512 acc_y = _mm512_setzero_ps();
+
+      for (int j = 0; j < n_p_pad; j += 16) {
+
+        // # 16 ADDs
+        __m512 j_idx = _mm512_add_ps(_mm512_set1_ps((float)j), lane_off);
+
+        __mmask16 m_neq_i = _mm512_cmp_ps_mask(j_idx, v_i, _CMP_NEQ_OQ);
+        __mmask16 m_in_range = _mm512_cmp_ps_mask(j_idx, v_np, _CMP_LT_OQ);
+        __mmask16 active = m_neq_i & m_in_range;
+
+        __m512 pjx = _mm512_loadu_ps(&pos_x[j]);
+        __m512 pjy = _mm512_loadu_ps(&pos_y[j]);
+        __m512 sbe_jx = _mm512_loadu_ps(&sb_e_x[j]);
+        __m512 sbe_jy = _mm512_loadu_ps(&sb_e_y[j]);
+        __m512 sb_jsq = _mm512_loadu_ps(&sb_sq[j]);
+
+        // # 32 ADDs
+        __m512 r_ab_x = _mm512_sub_ps(pix, pjx);
+        __m512 r_ab_y = _mm512_sub_ps(piy, pjy);
+
+        // # 32 MULTs, 16 ADDs
+        __m512 r_ab_n2 = _mm512_fmadd_ps(r_ab_x, r_ab_x, _mm512_mul_ps(r_ab_y, r_ab_y));
+
+        // # 16 SQRTs
+        __m512 r_ab_n = _mm512_sqrt_ps(r_ab_n2);
+
+        r_ab_n = _mm512_max_ps(r_ab_n, v_eps);
+
+        // # 32 ADDs
+        __m512 r_ab2_x = _mm512_sub_ps(r_ab_x, sbe_jx);
+        __m512 r_ab2_y = _mm512_sub_ps(r_ab_y, sbe_jy);
+
+        // # 32 MULTs, 16 ADDs
+        __m512 r_ab2_n2 = _mm512_fmadd_ps(r_ab2_x, r_ab2_x, _mm512_mul_ps(r_ab2_y, r_ab2_y));
+
+        // # 16 SQRTs
+        __m512 r_ab2_n = _mm512_sqrt_ps(r_ab2_n2);
+
+        r_ab2_n = _mm512_max_ps(r_ab2_n, v_eps);
+
+        // # 16 ADDs
+        __m512 r_sum = _mm512_add_ps(r_ab_n, r_ab2_n);
+
+        // # 16 MULTs, 16 ADDs
+        __m512 two_b_sq = _mm512_fmsub_ps(r_sum, r_sum, sb_jsq);
+
+        two_b_sq = _mm512_max_ps(two_b_sq, v_eps_sq);
+
+        // # 16 SQRTs
+        __m512 two_b = _mm512_sqrt_ps(two_b_sq);
+
+        // # 16 MULTs
+        __m512 e_arg = _mm512_mul_ps(two_b, v_neg_2s_inv);
+
+        e_arg = _mm512_max_ps(e_arg, v_exp_min);
+
+        // # 256 MULTs, 224 ADDs
+        __m512 e_val = _avx512_exp_ps(e_arg);
+
+        // # 64 MULTs
+        __m512 numer = _mm512_mul_ps(_mm512_mul_ps(r_sum, v_neg_v0_div_2s), e_val);
+        __m512 denom = _mm512_mul_ps(_mm512_mul_ps(two_b, r_ab_n), r_ab2_n);
+        __m512 denom_safe = _mm512_mask_blend_ps(active, v_one, denom);
+
+        // # 16 DIVs
+        __m512 smc = _mm512_div_ps(numer, denom_safe);
+
+        // # 32 MULTs
+        __m512 sm_r_ab = _mm512_mul_ps(smc, r_ab2_n);
+        __m512 sm_r_ab2 = _mm512_mul_ps(smc, r_ab_n);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 f_x = _mm512_fmadd_ps(r_ab2_x, sm_r_ab2, _mm512_mul_ps(r_ab_x, sm_r_ab));
+        __m512 f_y = _mm512_fmadd_ps(r_ab2_y, sm_r_ab2, _mm512_mul_ps(r_ab_y, sm_r_ab));
+
+        // # 32 MULTs, 16 ADDs
+        __m512 f_norm_sq = _mm512_fmadd_ps(f_x, f_x, _mm512_mul_ps(f_y, f_y));
+
+        // # 16 MULTs
+        __m512 rhs = _mm512_mul_ps(f_norm_sq, v_cos_phi_sq);
+
+        // # 32 MULTs, 16 ADDs
+        __m512 e_dot_f = _mm512_fmadd_ps(eix, f_x, _mm512_mul_ps(eiy, f_y));
+
+        // # 16 MULTs
+        __m512 e_dot_f_sq = _mm512_mul_ps(e_dot_f, e_dot_f);
+
+        __mmask16 c1 = _mm512_cmp_ps_mask(e_dot_f, v_zero, _CMP_LT_OQ);
+        __mmask16 c2 = _mm512_cmp_ps_mask(e_dot_f_sq, rhs, _CMP_GT_OQ);
+        __mmask16 cond = c1 & c2;
+
+        __m512 w_val = _mm512_mask_blend_ps(cond, v_dt_neg, v_c_neg_dt);
+
+        __m512 f_x_safe = _mm512_maskz_mov_ps(active, f_x);
+        __m512 f_y_safe = _mm512_maskz_mov_ps(active, f_y);
+
+        // # 32 MULTs, 32 ADDs
+        acc_x = _mm512_fmadd_ps(f_x_safe, w_val, acc_x);
+        acc_y = _mm512_fmadd_ps(f_y_safe, w_val, acc_y);
+      }
+
+      // # 32 ADDs
+      f_alpha_dt_x += _mm512_reduce_add_ps(acc_x);
+      f_alpha_dt_y += _mm512_reduce_add_ps(acc_y);
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
+        float v1_x = pos_x[i] - board.obstacles[k].start_x;
+        float v1_y = pos_y[i] - board.obstacles[k].start_y;
+        float v2_x = board.obstacles[k].obstacle_v_x;
+        float v2_y = board.obstacles[k].obstacle_v_y;
+
+        // # 3 MULTs, 1 ADD
+        float projection = (v1_x * v2_x + v1_y * v2_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0f)
+          projection = 0.0f;
+        else if (projection > 1.0f)
+          projection = 1.0f;
+
+        // # 4 ADDs, 2 MULTs
+        float r_aB_x = pos_x[i] - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        float r_aB_y = pos_y[i] - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        float r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+        r_aB_norm = r_aB_norm > EPS ? r_aB_norm : EPS;
+
+        // # 4 MULTs, 1 EXP, 1 DIV
+        float scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+        float f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        float f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        float w_val = dt_neg;
+        float e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -e_y[i] : e_y[i]);
+
+        if (e_dot_f < cos_phi)
+          w_val = c_neg_times_dt;
+
+        // # 2 MULTS
+        float f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        float f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        // # 2 ADDs
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      // # 2 ADDs
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p_pad; i += 16) {
+      __m512 w_x_i_vec = _mm512_loadu_ps(&w_x[i]);
+      __m512 w_y_i_vec = _mm512_loadu_ps(&w_y[i]);
+      __m512 v_max_i_vec = _mm512_loadu_ps(&v_max[i]);
+      __m512 pos_x_i_vec = _mm512_loadu_ps(&pos_x[i]);
+      __m512 pos_y_i_vec = _mm512_loadu_ps(&pos_y[i]);
+      __m512 destination_y_i_vec = _mm512_loadu_ps(&destination_y[i]);
+      __m512 destination_x_i_vec = _mm512_loadu_ps(&destination_x[i]);
+      __m512 desired_v0_i_vec = _mm512_loadu_ps(&desired_v0[i]);
+
+      __m512i color_i_vec = _mm512_loadu_si512((__m512i *)&color[i]);
+
+      // # 32 MULTs, 16 ADDs
+      __m512 w_norm_sq = _mm512_fmadd_ps(w_x_i_vec, w_x_i_vec, _mm512_mul_ps(w_y_i_vec, w_y_i_vec));
+
+      // # 16 SQRTs
+      __m512 w_norm = _mm512_sqrt_ps(w_norm_sq);
+      __m512 w_norm_safe = _mm512_max_ps(w_norm, _mm512_set1_ps(EPS));
+
+      // # 16 DIVs
+      __m512 correction_factor = _mm512_min_ps(_mm512_set1_ps(1.0f), _mm512_div_ps(v_max_i_vec, w_norm_safe));
+
+      // # 32 MULTs
+      __m512 curr_vel_x_i_vec = _mm512_mul_ps(w_x_i_vec, correction_factor);
+      __m512 curr_vel_y_i_vec = _mm512_mul_ps(w_y_i_vec, correction_factor);
+
+      __m512 dt_vec = _mm512_set1_ps(dt);
+
+      // # 32 MULTs, 32 ADDs
+      pos_x_i_vec = _mm512_fmadd_ps(curr_vel_x_i_vec, dt_vec, pos_x_i_vec);
+      pos_y_i_vec = _mm512_fmadd_ps(curr_vel_y_i_vec, dt_vec, pos_y_i_vec);
+
+      __mmask16 cond_1_left = _mm512_cmpeq_epi32_mask(color_i_vec, _mm512_setzero_si512());
+      __mmask16 cond_1_right = _mm512_cmp_ps_mask(pos_x_i_vec, _mm512_set1_ps(BOARD_LENGTH_MINUS_ONE_POINT_FIVE), _CMP_GT_OQ);
+      __mmask16 cond_1 = cond_1_left & cond_1_right;
+
+      pos_x_i_vec = _mm512_mask_blend_ps(cond_1, pos_x_i_vec, _mm512_set1_ps(1.0f));
+      destination_y_i_vec = _mm512_mask_blend_ps(cond_1, destination_y_i_vec, pos_y_i_vec);
+      w_x_i_vec = _mm512_mask_blend_ps(cond_1, w_x_i_vec, desired_v0_i_vec);
+      w_y_i_vec = _mm512_mask_blend_ps(cond_1, w_y_i_vec, _mm512_setzero_ps());
+      curr_vel_x_i_vec = _mm512_mask_blend_ps(cond_1, curr_vel_x_i_vec, w_x_i_vec);
+      curr_vel_y_i_vec = _mm512_mask_blend_ps(cond_1, curr_vel_y_i_vec, w_y_i_vec);
+
+      __mmask16 cond_2_left = _mm512_cmpeq_epi32_mask(color_i_vec, _mm512_set1_epi32(1));
+      __mmask16 cond_2_right = _mm512_cmp_ps_mask(pos_x_i_vec, _mm512_set1_ps(1.5f), _CMP_LT_OQ);
+      __mmask16 cond_2 = cond_2_left & cond_2_right;
+
+      pos_x_i_vec = _mm512_mask_blend_ps(cond_2, pos_x_i_vec, _mm512_set1_ps(BOARD_LENGTH_MINUS_ONE));
+      destination_y_i_vec = _mm512_mask_blend_ps(cond_2, destination_y_i_vec, pos_y_i_vec);
+      __m512 neg_desired_v0 = _mm512_xor_ps(desired_v0_i_vec, _mm512_set1_ps(-0.0f));
+      w_x_i_vec = _mm512_mask_blend_ps(cond_2, w_x_i_vec, neg_desired_v0);
+      w_y_i_vec = _mm512_mask_blend_ps(cond_2, w_y_i_vec, _mm512_setzero_ps());
+      curr_vel_x_i_vec = _mm512_mask_blend_ps(cond_2, curr_vel_x_i_vec, w_x_i_vec);
+      curr_vel_y_i_vec = _mm512_mask_blend_ps(cond_2, curr_vel_y_i_vec, w_y_i_vec);
+
+      pos_y_i_vec = _mm512_max_ps(pos_y_i_vec, _mm512_set1_ps(WALL_EPS));
+      pos_y_i_vec = _mm512_min_ps(pos_y_i_vec, _mm512_set1_ps(BOARD_WIDTH_MINUS_WALL_EPS));
+
+      // # 32 ADDs
+      __m512 ex_vec = _mm512_sub_ps(destination_x_i_vec, pos_x_i_vec);
+      __m512 ey_vec = _mm512_sub_ps(destination_y_i_vec, pos_y_i_vec);
+
+      // # 32 MULTs, 16 ADDs
+      __m512 e_norm_sq = _mm512_fmadd_ps(ex_vec, ex_vec, _mm512_mul_ps(ey_vec, ey_vec));
+
+      // # 16 SQRTs
+      __m512 e_norm = _mm512_sqrt_ps(e_norm_sq);
+
+      // # 16 DIVs
+      __m512 e_norm_inv = _mm512_div_ps(_mm512_set1_ps(1.0f), e_norm);
+
+      // # 32 MULTs
+      __m512 e_x_i_vec = _mm512_mul_ps(ex_vec, e_norm_inv);
+      __m512 e_y_i_vec = _mm512_mul_ps(ey_vec, e_norm_inv);
+
+      // # 32 MULTs, 16 ADDs
+      __m512 vel_norm_sq = _mm512_fmadd_ps(curr_vel_x_i_vec, curr_vel_x_i_vec, _mm512_mul_ps(curr_vel_y_i_vec, curr_vel_y_i_vec));
+
+      // # 16 MULTs
+      __m512 sb_sq_i_vec = _mm512_mul_ps(vel_norm_sq, _mm512_set1_ps(delta_t_sq));
+
+      // # 16 SQRTs
+      __m512 vel_norm_dt = _mm512_sqrt_ps(sb_sq_i_vec);
+
+      _mm512_storeu_ps(&curr_vel_x[i], curr_vel_x_i_vec);
+      _mm512_storeu_ps(&curr_vel_y[i], curr_vel_y_i_vec);
+      _mm512_storeu_ps(&w_x[i], w_x_i_vec);
+      _mm512_storeu_ps(&w_y[i], w_y_i_vec);
+      _mm512_storeu_ps(&pos_x[i], pos_x_i_vec);
+      _mm512_storeu_ps(&pos_y[i], pos_y_i_vec);
+      _mm512_storeu_ps(&e_x[i], e_x_i_vec);
+      _mm512_storeu_ps(&e_y[i], e_y_i_vec);
+      _mm512_storeu_ps(&destination_y[i], destination_y_i_vec);
+      _mm512_storeu_ps(&sb_sq[i], sb_sq_i_vec);
+
+      // # 32 MULTs
+      _mm512_storeu_ps(&sb_e_x[i], _mm512_mul_ps(e_x_i_vec, vel_norm_dt));
+      _mm512_storeu_ps(&sb_e_y[i], _mm512_mul_ps(e_y_i_vec, vel_norm_dt));
+    }
+
+#if !BENCHMARK
+    for (int i = 0; i < n_p; i++) {
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+    }
+#endif
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp
new file mode 100644
index 0000000..85eac51
--- /dev/null
+++ b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp
@@ -0,0 +1,638 @@
+// # Changes: Added ILP with k = 2 to j-loop, removed EPS checks where possible
+// # Predecessor: variant_float_inlining_SoA_AVX512_polyexp_moreavx.cpp
+
+#include "../headers/board_reduced_float.hpp"
+#include "../headers/constants_float.hpp"
+#include "../headers/pedestrian_reduced_float.hpp"
+
+#include <immintrin.h>
+
+#include <fstream>
+#include <random>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0f
+#define BOARD_LENGTH 50.0f
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0f
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5f
+#define WALL_EPS 1e-3f
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999f
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+alignas(64) vector<float> pos_x, pos_y;
+alignas(64) vector<float> sb_sq;
+alignas(64) vector<float> sb_e_x, sb_e_y;
+alignas(64) vector<float> e_x, e_y;
+alignas(64) vector<float> curr_vel_x, curr_vel_y;
+alignas(64) vector<float> w_x, w_y;
+alignas(64) vector<float> desired_v0;
+alignas(64) vector<float> v_max;
+alignas(64) vector<float> destination_x, destination_y;
+alignas(64) vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<float> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<float> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<float> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<float> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  const int n_p_pad = (n_p + 31) & ~31;
+
+  pos_x.resize(n_p_pad);
+  pos_y.resize(n_p_pad);
+  sb_sq.resize(n_p_pad);
+  sb_e_x.resize(n_p_pad);
+  sb_e_y.resize(n_p_pad);
+  e_x.resize(n_p_pad);
+  e_y.resize(n_p_pad);
+  curr_vel_x.resize(n_p_pad);
+  curr_vel_y.resize(n_p_pad);
+  w_x.resize(n_p_pad);
+  w_y.resize(n_p_pad);
+  desired_v0.resize(n_p_pad);
+  v_max.resize(n_p_pad);
+  destination_x.resize(n_p_pad);
+  destination_y.resize(n_p_pad);
+  color.resize(n_p_pad);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    float x = x_gen_left(gen);
+    float y = y_gen(gen);
+
+    float dest_x = BOARD_LENGTH + 30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    float x = x_gen_right(gen);
+    float y = y_gen(gen);
+
+    float dest_x = -30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int j = n_p; j < n_p_pad; j++) {
+    pos_x[j] = 1e9f;
+    pos_y[j] = 1e9f;
+    sb_sq[j] = 0.0f;
+    sb_e_x[j] = 0.0f;
+    sb_e_y[j] = 0.0f;
+  }
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+// # 256 MULTs, 224 ADDs
+static inline __m512 _avx512_exp_ps(__m512 x) {
+
+  const __m512 log2e = _mm512_set1_ps(1.4426950408889634f);
+
+  // # 16 MULTs
+  __m512 nf = _mm512_roundscale_ps(_mm512_mul_ps(x, log2e), _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+
+  const __m512 ln2 = _mm512_set1_ps(0.6931471805599453f);
+
+  // # 16 MULTs, 16 ADDs
+  __m512 r = _mm512_fnmadd_ps(nf, ln2, x);
+
+  const __m512 c13 = _mm512_set1_ps(c13_scalar);
+  const __m512 c12 = _mm512_set1_ps(c12_scalar);
+  const __m512 c11 = _mm512_set1_ps(c11_scalar);
+  const __m512 c10 = _mm512_set1_ps(c10_scalar);
+  const __m512 c9 = _mm512_set1_ps(c9_scalar);
+  const __m512 c8 = _mm512_set1_ps(c8_scalar);
+  const __m512 c7 = _mm512_set1_ps(c7_scalar);
+  const __m512 c6 = _mm512_set1_ps(c6_scalar);
+  const __m512 c5 = _mm512_set1_ps(c5_scalar);
+  const __m512 c4 = _mm512_set1_ps(c4_scalar);
+  const __m512 c3 = _mm512_set1_ps(c3_scalar);
+  const __m512 c2 = _mm512_set1_ps(c2_scalar);
+  const __m512 c1 = _mm512_set1_ps(c1_scalar);
+  const __m512 c0 = _mm512_set1_ps(c0_scalar);
+
+  __m512 p = c13;
+
+  // #   13 * (16 MULTs, 16 ADDs)
+  // # = 208 MULTs, 208 ADDs
+  p = _mm512_fmadd_ps(p, r, c12);
+  p = _mm512_fmadd_ps(p, r, c11);
+  p = _mm512_fmadd_ps(p, r, c10);
+  p = _mm512_fmadd_ps(p, r, c9);
+  p = _mm512_fmadd_ps(p, r, c8);
+  p = _mm512_fmadd_ps(p, r, c7);
+  p = _mm512_fmadd_ps(p, r, c6);
+  p = _mm512_fmadd_ps(p, r, c5);
+  p = _mm512_fmadd_ps(p, r, c4);
+  p = _mm512_fmadd_ps(p, r, c3);
+  p = _mm512_fmadd_ps(p, r, c2);
+  p = _mm512_fmadd_ps(p, r, c1);
+  p = _mm512_fmadd_ps(p, r, c0);
+
+  __m512i ni = _mm512_cvtps_epi32(nf);
+  __m512i biased = _mm512_add_epi32(ni, _mm512_set1_epi32(127));
+  __m512i scaled = _mm512_slli_epi32(biased, 23);
+  __m512 pow2n = _mm512_castsi512_ps(scaled);
+
+  // # 16 MULTs
+  return _mm512_mul_ps(p, pow2n);
+}
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+  (void)pedestrians;
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+  vector<tuple<int, int, float, float, float, float, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+    for (int i = 0; i < n_p; i++) {
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  const int n_p_pad = (n_p + 31) & ~31;
+
+  alignas(64) const float lane_offset_arr[16] = {0.0f, 1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f};
+  const __m512 lane_off = _mm512_load_ps(lane_offset_arr);
+
+  const __m512 v_neg_v0_div_2s = _mm512_set1_ps(NEG_V0_AB_div_TWO_SIGMA);
+  const __m512 v_neg_2s_inv = _mm512_set1_ps(NEG_TWO_SIGMA_INV);
+  const __m512 v_cos_phi_sq = _mm512_set1_ps(cos_phi_sq);
+  const __m512 v_dt_neg = _mm512_set1_ps(dt_neg);
+  const __m512 v_c_neg_dt = _mm512_set1_ps(c_neg_times_dt);
+  const __m512 v_zero = _mm512_setzero_ps();
+  const __m512 v_one = _mm512_set1_ps(1.0f);
+  const __m512 v_eps = _mm512_set1_ps(EPS);
+  const __m512 v_eps_sq = _mm512_set1_ps(EPS_SQ);
+  const __m512 v_np = _mm512_set1_ps((float)n_p);
+  const __m512 v_exp_min = _mm512_set1_ps(-87.0f);
+
+  // # ADDs: n_iter * (
+  // #   n_p * 68 +
+  // #   n_p * n_p_pad * 29 +
+  // #   n_p * board.n_obstacles * 10 +
+  // #   n_p_pad * 7
+  // # )
+
+  // # MULTs: n_iter * (
+  // #   n_p * 4 +
+  // #   n_p * n_p_pad * 40 +
+  // #   n_p * board.n_obstacles * 13 +
+  // #   n_p_pad * 15
+  // # )
+
+  // # SQRTs: n_iter * (
+  // #   n_p * n_p_pad * 3 +
+  // #   n_p * board.n_obstacles +
+  // #   n_p_pad * 3
+  // # )
+
+  // # DIVs: n_iter * (
+  // #   n_p * n_p_pad +
+  // #   n_p * board.n_obstacles +
+  // #   n_p_pad * 2
+  // # )
+
+  // # EXPs: n_iter * n_p * board.n_obstacles
+
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+
+      // # 2 MULTs
+      float desired_velocity_x = e_x[i] * desired_v0[i];
+      float desired_velocity_y = e_y[i] * desired_v0[i];
+
+      // # 2 ADDs, 2 MULTs
+      float f_alpha_dt_x = (desired_velocity_x - curr_vel_x[i]) * dt_times_tau_alpha_inv;
+      float f_alpha_dt_y = (desired_velocity_y - curr_vel_y[i]) * dt_times_tau_alpha_inv;
+
+      const __m512 pix = _mm512_set1_ps(pos_x[i]);
+      const __m512 piy = _mm512_set1_ps(pos_y[i]);
+      const __m512 eix = _mm512_set1_ps(e_x[i]);
+      const __m512 eiy = _mm512_set1_ps(e_y[i]);
+      const __m512 v_i = _mm512_set1_ps((float)i);
+
+      __m512 acc_x_0 = _mm512_setzero_ps();
+      __m512 acc_y_0 = _mm512_setzero_ps();
+      __m512 acc_x_1 = _mm512_setzero_ps();
+      __m512 acc_y_1 = _mm512_setzero_ps();
+
+      for (int j = 0; j < n_p_pad; j += 32) {
+
+        // # 32 ADDs
+        __m512 j_idx_0 = _mm512_add_ps(_mm512_set1_ps((float)j), lane_off);
+        __m512 j_idx_1 = _mm512_add_ps(_mm512_set1_ps((float)j + 16), lane_off);
+
+        __mmask16 m_neq_i_0 = _mm512_cmp_ps_mask(j_idx_0, v_i, _CMP_NEQ_OQ);
+        __mmask16 m_neq_i_1 = _mm512_cmp_ps_mask(j_idx_1, v_i, _CMP_NEQ_OQ);
+
+        __mmask16 m_in_range_0 = _mm512_cmp_ps_mask(j_idx_0, v_np, _CMP_LT_OQ);
+        __mmask16 m_in_range_1 = _mm512_cmp_ps_mask(j_idx_1, v_np, _CMP_LT_OQ);
+
+        __mmask16 active_0 = m_neq_i_0 & m_in_range_0;
+        __mmask16 active_1 = m_neq_i_1 & m_in_range_1;
+
+        __m512 pjx_0 = _mm512_loadu_ps(&pos_x[j]);
+        __m512 pjx_1 = _mm512_loadu_ps(&pos_x[j + 16]);
+
+        __m512 pjy_0 = _mm512_loadu_ps(&pos_y[j]);
+        __m512 pjy_1 = _mm512_loadu_ps(&pos_y[j + 16]);
+
+        __m512 sbe_jx_0 = _mm512_loadu_ps(&sb_e_x[j]);
+        __m512 sbe_jx_1 = _mm512_loadu_ps(&sb_e_x[j + 16]);
+
+        __m512 sbe_jy_0 = _mm512_loadu_ps(&sb_e_y[j]);
+        __m512 sbe_jy_1 = _mm512_loadu_ps(&sb_e_y[j + 16]);
+
+        __m512 sb_jsq_0 = _mm512_loadu_ps(&sb_sq[j]);
+        __m512 sb_jsq_1 = _mm512_loadu_ps(&sb_sq[j + 16]);
+
+        // # 32 ADDs
+        __m512 r_ab_x_0 = _mm512_sub_ps(pix, pjx_0);
+        __m512 r_ab_x_1 = _mm512_sub_ps(pix, pjx_1);
+
+        // # 32 ADDs
+        __m512 r_ab_y_0 = _mm512_sub_ps(piy, pjy_0);
+        __m512 r_ab_y_1 = _mm512_sub_ps(piy, pjy_1);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 r_ab_n2_0 = _mm512_fmadd_ps(r_ab_x_0, r_ab_x_0, _mm512_mul_ps(r_ab_y_0, r_ab_y_0));
+        __m512 r_ab_n2_1 = _mm512_fmadd_ps(r_ab_x_1, r_ab_x_1, _mm512_mul_ps(r_ab_y_1, r_ab_y_1));
+
+        // # 32 SQRTs
+        __m512 r_ab_n_0 = _mm512_sqrt_ps(r_ab_n2_0);
+        __m512 r_ab_n_1 = _mm512_sqrt_ps(r_ab_n2_1);
+
+        // # 32 ADDs
+        __m512 r_ab2_x_0 = _mm512_sub_ps(r_ab_x_0, sbe_jx_0);
+        __m512 r_ab2_x_1 = _mm512_sub_ps(r_ab_x_1, sbe_jx_1);
+
+        // # 32 ADDs
+        __m512 r_ab2_y_0 = _mm512_sub_ps(r_ab_y_0, sbe_jy_0);
+        __m512 r_ab2_y_1 = _mm512_sub_ps(r_ab_y_1, sbe_jy_1);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 r_ab2_n2_0 = _mm512_fmadd_ps(r_ab2_x_0, r_ab2_x_0, _mm512_mul_ps(r_ab2_y_0, r_ab2_y_0));
+        __m512 r_ab2_n2_1 = _mm512_fmadd_ps(r_ab2_x_1, r_ab2_x_1, _mm512_mul_ps(r_ab2_y_1, r_ab2_y_1));
+
+        // # 32 SQRTs
+        __m512 r_ab2_n_0 = _mm512_sqrt_ps(r_ab2_n2_0);
+        __m512 r_ab2_n_1 = _mm512_sqrt_ps(r_ab2_n2_1);
+
+        // # 32 ADDs
+        __m512 r_sum_0 = _mm512_add_ps(r_ab_n_0, r_ab2_n_0);
+        __m512 r_sum_1 = _mm512_add_ps(r_ab_n_1, r_ab2_n_1);
+
+        // # 32 MULTs, 32 ADDs
+        __m512 two_b_sq_0 = _mm512_fmsub_ps(r_sum_0, r_sum_0, sb_jsq_0);
+        __m512 two_b_sq_1 = _mm512_fmsub_ps(r_sum_1, r_sum_1, sb_jsq_1);
+
+        // # 32 SQRTs
+        __m512 two_b_0 = _mm512_sqrt_ps(two_b_sq_0);
+        __m512 two_b_1 = _mm512_sqrt_ps(two_b_sq_1);
+
+        // # 32 MULTs
+        __m512 e_arg_0 = _mm512_mul_ps(two_b_0, v_neg_2s_inv);
+        __m512 e_arg_1 = _mm512_mul_ps(two_b_1, v_neg_2s_inv);
+
+        e_arg_0 = _mm512_max_ps(e_arg_0, v_exp_min);
+        e_arg_1 = _mm512_max_ps(e_arg_1, v_exp_min);
+
+        // # 512 MULTs, 448 ADDs
+        __m512 e_val_0 = _avx512_exp_ps(e_arg_0);
+        __m512 e_val_1 = _avx512_exp_ps(e_arg_1);
+
+        // # 64 MULTs
+        __m512 numer_0 = _mm512_mul_ps(_mm512_mul_ps(r_sum_0, v_neg_v0_div_2s), e_val_0);
+        __m512 numer_1 = _mm512_mul_ps(_mm512_mul_ps(r_sum_1, v_neg_v0_div_2s), e_val_1);
+
+        // # 64 MULTs
+        __m512 denom_0 = _mm512_mul_ps(_mm512_mul_ps(two_b_0, r_ab_n_0), r_ab2_n_0);
+        __m512 denom_1 = _mm512_mul_ps(_mm512_mul_ps(two_b_1, r_ab_n_1), r_ab2_n_1);
+
+        __m512 denom_safe_0 = _mm512_mask_blend_ps(active_0, v_one, denom_0);
+        __m512 denom_safe_1 = _mm512_mask_blend_ps(active_1, v_one, denom_1);
+
+        // # 32 DIVs
+        __m512 smc_0 = _mm512_div_ps(numer_0, denom_safe_0);
+        __m512 smc_1 = _mm512_div_ps(numer_1, denom_safe_1);
+
+        // # 32 MULTs
+        __m512 sm_r_ab_0 = _mm512_mul_ps(smc_0, r_ab2_n_0);
+        __m512 sm_r_ab_1 = _mm512_mul_ps(smc_1, r_ab2_n_1);
+
+        // # 32 MULTs
+        __m512 sm_r_ab2_0 = _mm512_mul_ps(smc_0, r_ab_n_0);
+        __m512 sm_r_ab2_1 = _mm512_mul_ps(smc_1, r_ab_n_1);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 f_x_0 = _mm512_fmadd_ps(r_ab2_x_0, sm_r_ab2_0, _mm512_mul_ps(r_ab_x_0, sm_r_ab_0));
+        __m512 f_x_1 = _mm512_fmadd_ps(r_ab2_x_1, sm_r_ab2_1, _mm512_mul_ps(r_ab_x_1, sm_r_ab_1));
+
+        // # 64 MULTs, 32 ADDs
+        __m512 f_y_0 = _mm512_fmadd_ps(r_ab2_y_0, sm_r_ab2_0, _mm512_mul_ps(r_ab_y_0, sm_r_ab_0));
+        __m512 f_y_1 = _mm512_fmadd_ps(r_ab2_y_1, sm_r_ab2_1, _mm512_mul_ps(r_ab_y_1, sm_r_ab_1));
+
+        // # 64 MULTs, 32 ADDs
+        __m512 f_norm_sq_0 = _mm512_fmadd_ps(f_x_0, f_x_0, _mm512_mul_ps(f_y_0, f_y_0));
+        __m512 f_norm_sq_1 = _mm512_fmadd_ps(f_x_1, f_x_1, _mm512_mul_ps(f_y_1, f_y_1));
+
+        // # 32 MULTs
+        __m512 rhs_0 = _mm512_mul_ps(f_norm_sq_0, v_cos_phi_sq);
+        __m512 rhs_1 = _mm512_mul_ps(f_norm_sq_1, v_cos_phi_sq);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 e_dot_f_0 = _mm512_fmadd_ps(eix, f_x_0, _mm512_mul_ps(eiy, f_y_0));
+        __m512 e_dot_f_1 = _mm512_fmadd_ps(eix, f_x_1, _mm512_mul_ps(eiy, f_y_1));
+
+        // # 32 MULTs
+        __m512 e_dot_f_sq_0 = _mm512_mul_ps(e_dot_f_0, e_dot_f_0);
+        __m512 e_dot_f_sq_1 = _mm512_mul_ps(e_dot_f_1, e_dot_f_1);
+
+        __mmask16 c1_0 = _mm512_cmp_ps_mask(e_dot_f_0, v_zero, _CMP_LT_OQ);
+        __mmask16 c1_1 = _mm512_cmp_ps_mask(e_dot_f_1, v_zero, _CMP_LT_OQ);
+
+        __mmask16 c2_0 = _mm512_cmp_ps_mask(e_dot_f_sq_0, rhs_0, _CMP_GT_OQ);
+        __mmask16 c2_1 = _mm512_cmp_ps_mask(e_dot_f_sq_1, rhs_1, _CMP_GT_OQ);
+
+        __mmask16 cond_0 = c1_0 & c2_0;
+        __mmask16 cond_1 = c1_1 & c2_1;
+
+        __m512 w_val_0 = _mm512_mask_blend_ps(cond_0, v_dt_neg, v_c_neg_dt);
+        __m512 w_val_1 = _mm512_mask_blend_ps(cond_1, v_dt_neg, v_c_neg_dt);
+
+        __m512 f_x_safe_0 = _mm512_maskz_mov_ps(active_0, f_x_0);
+        __m512 f_x_safe_1 = _mm512_maskz_mov_ps(active_1, f_x_1);
+
+        __m512 f_y_safe_0 = _mm512_maskz_mov_ps(active_0, f_y_0);
+        __m512 f_y_safe_1 = _mm512_maskz_mov_ps(active_1, f_y_1);
+
+        // # 32 MULTs, 32 ADDs
+        acc_x_0 = _mm512_fmadd_ps(f_x_safe_0, w_val_0, acc_x_0);
+        acc_x_1 = _mm512_fmadd_ps(f_x_safe_1, w_val_1, acc_x_1);
+
+        // # 32 MULTs, 32 ADDs
+        acc_y_0 = _mm512_fmadd_ps(f_y_safe_0, w_val_0, acc_y_0);
+        acc_y_1 = _mm512_fmadd_ps(f_y_safe_1, w_val_1, acc_y_1);
+      }
+
+      // # 64 ADDs
+      f_alpha_dt_x += _mm512_reduce_add_ps(_mm512_add_ps(acc_x_0, acc_x_1));
+      f_alpha_dt_y += _mm512_reduce_add_ps(_mm512_add_ps(acc_y_0, acc_y_1));
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
+        float v1_x = pos_x[i] - board.obstacles[k].start_x;
+        float v1_y = pos_y[i] - board.obstacles[k].start_y;
+        float v2_x = board.obstacles[k].obstacle_v_x;
+        float v2_y = board.obstacles[k].obstacle_v_y;
+
+        // # 3 MULTs, 1 ADD
+        float projection = (v1_x * v2_x + v1_y * v2_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0f)
+          projection = 0.0f;
+        else if (projection > 1.0f)
+          projection = 1.0f;
+
+        // # 4 ADDs, 2 MULTs
+        float r_aB_x = pos_x[i] - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        float r_aB_y = pos_y[i] - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        float r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+        r_aB_norm = r_aB_norm > EPS ? r_aB_norm : EPS;
+
+        // # 1 EXP, 4 MULTs, 1 DIV
+        float scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+        float f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        float f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        float w_val = dt_neg;
+        float e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -e_y[i] : e_y[i]);
+
+        if (e_dot_f < cos_phi)
+          w_val = c_neg_times_dt;
+
+        // # 2 MULTs
+        float f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        float f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        // # 2 ADDs
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      // # 2 ADDs
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p_pad; i += 16) {
+      
+      __m512 w_x_i_vec = _mm512_loadu_ps(&w_x[i]);
+      __m512 w_y_i_vec = _mm512_loadu_ps(&w_y[i]);
+      __m512 v_max_i_vec = _mm512_loadu_ps(&v_max[i]);
+      __m512 pos_x_i_vec = _mm512_loadu_ps(&pos_x[i]);
+      __m512 pos_y_i_vec = _mm512_loadu_ps(&pos_y[i]);
+      __m512 destination_y_i_vec = _mm512_loadu_ps(&destination_y[i]);
+      __m512 destination_x_i_vec = _mm512_loadu_ps(&destination_x[i]);
+      __m512 desired_v0_i_vec = _mm512_loadu_ps(&desired_v0[i]);
+      
+      __m512i color_i_vec = _mm512_loadu_si512((__m512i*)&color[i]);
+
+      // # 32 MULTs, 16 ADDs
+      __m512 w_norm_sq = _mm512_fmadd_ps(w_x_i_vec, w_x_i_vec, _mm512_mul_ps(w_y_i_vec, w_y_i_vec));
+
+      // # 16 SQRTs
+      __m512 w_norm = _mm512_sqrt_ps(w_norm_sq);
+
+      // # 16 DIVs
+      __m512 correction_factor = _mm512_min_ps(_mm512_set1_ps(1.0f), _mm512_div_ps(v_max_i_vec, w_norm));
+
+      // # 32 MULTs
+      __m512 curr_vel_x_i_vec = _mm512_mul_ps(w_x_i_vec, correction_factor);
+      __m512 curr_vel_y_i_vec = _mm512_mul_ps(w_y_i_vec, correction_factor);
+
+      __m512 dt_vec = _mm512_set1_ps(dt);
+
+      // # 32 MULTs, 32 ADDs
+      pos_x_i_vec = _mm512_fmadd_ps(curr_vel_x_i_vec, dt_vec, pos_x_i_vec);
+      pos_y_i_vec = _mm512_fmadd_ps(curr_vel_y_i_vec, dt_vec, pos_y_i_vec);
+
+      __mmask16 cond_1_left = _mm512_cmpeq_epi32_mask(color_i_vec, _mm512_setzero_si512());
+      __mmask16 cond_1_right = _mm512_cmp_ps_mask(pos_x_i_vec, _mm512_set1_ps(BOARD_LENGTH_MINUS_ONE_POINT_FIVE), _CMP_GT_OQ);
+      __mmask16 cond_1 = cond_1_left & cond_1_right;
+
+      pos_x_i_vec = _mm512_mask_blend_ps(cond_1, pos_x_i_vec, _mm512_set1_ps(1.0f));
+      destination_y_i_vec = _mm512_mask_blend_ps(cond_1, destination_y_i_vec, pos_y_i_vec);
+      w_x_i_vec = _mm512_mask_blend_ps(cond_1, w_x_i_vec, desired_v0_i_vec);
+      w_y_i_vec = _mm512_mask_blend_ps(cond_1, w_y_i_vec, _mm512_setzero_ps());
+      curr_vel_x_i_vec = _mm512_mask_blend_ps(cond_1, curr_vel_x_i_vec, w_x_i_vec);
+      curr_vel_y_i_vec = _mm512_mask_blend_ps(cond_1, curr_vel_y_i_vec, w_y_i_vec);
+
+      __mmask16 cond_2_left = _mm512_cmpeq_epi32_mask(color_i_vec, _mm512_set1_epi32(1));
+      __mmask16 cond_2_right = _mm512_cmp_ps_mask(pos_x_i_vec, _mm512_set1_ps(1.5f), _CMP_LT_OQ);
+      __mmask16 cond_2 = cond_2_left & cond_2_right;
+
+      pos_x_i_vec = _mm512_mask_blend_ps(cond_2, pos_x_i_vec, _mm512_set1_ps(BOARD_LENGTH_MINUS_ONE));
+      destination_y_i_vec = _mm512_mask_blend_ps(cond_2, destination_y_i_vec, pos_y_i_vec);
+      __m512 neg_desired_v0 = _mm512_xor_ps(desired_v0_i_vec, _mm512_set1_ps(-0.0f));
+      w_x_i_vec = _mm512_mask_blend_ps(cond_2, w_x_i_vec, neg_desired_v0);
+      w_y_i_vec = _mm512_mask_blend_ps(cond_2, w_y_i_vec, _mm512_setzero_ps());
+      curr_vel_x_i_vec = _mm512_mask_blend_ps(cond_2, curr_vel_x_i_vec, w_x_i_vec);
+      curr_vel_y_i_vec = _mm512_mask_blend_ps(cond_2, curr_vel_y_i_vec, w_y_i_vec);
+
+      pos_y_i_vec = _mm512_max_ps(pos_y_i_vec, _mm512_set1_ps(WALL_EPS));
+      pos_y_i_vec = _mm512_min_ps(pos_y_i_vec, _mm512_set1_ps(BOARD_WIDTH_MINUS_WALL_EPS));
+
+      // # 32 ADDs
+      __m512 ex_vec = _mm512_sub_ps(destination_x_i_vec, pos_x_i_vec);
+      __m512 ey_vec = _mm512_sub_ps(destination_y_i_vec, pos_y_i_vec);
+
+      // # 32 MULTs, 16 ADDs
+      __m512 e_norm_sq = _mm512_fmadd_ps(ex_vec, ex_vec, _mm512_mul_ps(ey_vec, ey_vec));
+
+      // # 16 SQRTs
+      __m512 e_norm = _mm512_sqrt_ps(e_norm_sq);
+
+      // # 16 DIVs
+      __m512 e_norm_inv = _mm512_div_ps(_mm512_set1_ps(1.0f), e_norm);
+
+      // # 32 MULTs
+      __m512 e_x_i_vec = _mm512_mul_ps(ex_vec, e_norm_inv);
+      __m512 e_y_i_vec = _mm512_mul_ps(ey_vec, e_norm_inv);
+
+      // # 32 MULTs, 16 ADDs
+      __m512 vel_norm_sq = _mm512_fmadd_ps(curr_vel_x_i_vec, curr_vel_x_i_vec, _mm512_mul_ps(curr_vel_y_i_vec, curr_vel_y_i_vec));
+
+      // # 16 MULTs
+      __m512 sb_sq_i_vec = _mm512_mul_ps(vel_norm_sq, _mm512_set1_ps(delta_t_sq));
+
+      // # 16 SQRTs
+      __m512 vel_norm_dt = _mm512_sqrt_ps(sb_sq_i_vec);
+
+      _mm512_storeu_ps(&curr_vel_x[i], curr_vel_x_i_vec);
+      _mm512_storeu_ps(&curr_vel_y[i], curr_vel_y_i_vec);
+      _mm512_storeu_ps(&w_x[i], w_x_i_vec);
+      _mm512_storeu_ps(&w_y[i], w_y_i_vec);
+      _mm512_storeu_ps(&pos_x[i], pos_x_i_vec);
+      _mm512_storeu_ps(&pos_y[i], pos_y_i_vec);
+      _mm512_storeu_ps(&e_x[i], e_x_i_vec);
+      _mm512_storeu_ps(&e_y[i], e_y_i_vec);
+      _mm512_storeu_ps(&destination_y[i], destination_y_i_vec);
+      _mm512_storeu_ps(&sb_sq[i], sb_sq_i_vec);
+
+      // # 32 MULTs
+      _mm512_storeu_ps(&sb_e_x[i], _mm512_mul_ps(e_x_i_vec, vel_norm_dt));
+      _mm512_storeu_ps(&sb_e_y[i], _mm512_mul_ps(e_y_i_vec, vel_norm_dt));
+    }
+
+#if !BENCHMARK
+    for (int i = 0; i < n_p; i++) {
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+    }
+#endif
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix.cpp b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix.cpp
new file mode 100644
index 0000000..25a468c
--- /dev/null
+++ b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_moreilp_expfix.cpp
@@ -0,0 +1,735 @@
+// # Changes: Added ILP with k = 2 to second pass
+// # Predecessor: variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2.cpp
+
+#include "../headers/board_reduced_float.hpp"
+#include "../headers/constants_float.hpp"
+#include "../headers/pedestrian_reduced_float.hpp"
+
+#include <immintrin.h>
+
+#include <fstream>
+#include <random>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0f
+#define BOARD_LENGTH 50.0f
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0f
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5f
+#define WALL_EPS 1e-3f
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999f
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+alignas(64) vector<float> pos_x, pos_y;
+alignas(64) vector<float> sb_sq;
+alignas(64) vector<float> sb_e_x, sb_e_y;
+alignas(64) vector<float> e_x, e_y;
+alignas(64) vector<float> curr_vel_x, curr_vel_y;
+alignas(64) vector<float> w_x, w_y;
+alignas(64) vector<float> desired_v0;
+alignas(64) vector<float> v_max;
+alignas(64) vector<float> destination_x, destination_y;
+alignas(64) vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<float> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<float> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<float> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<float> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  const int n_p_pad = (n_p + 31) & ~31;
+
+  pos_x.resize(n_p_pad);
+  pos_y.resize(n_p_pad);
+  sb_sq.resize(n_p_pad);
+  sb_e_x.resize(n_p_pad);
+  sb_e_y.resize(n_p_pad);
+  e_x.resize(n_p_pad);
+  e_y.resize(n_p_pad);
+  curr_vel_x.resize(n_p_pad);
+  curr_vel_y.resize(n_p_pad);
+  w_x.resize(n_p_pad);
+  w_y.resize(n_p_pad);
+  desired_v0.resize(n_p_pad);
+  v_max.resize(n_p_pad);
+  destination_x.resize(n_p_pad);
+  destination_y.resize(n_p_pad);
+  color.resize(n_p_pad);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    float x = x_gen_left(gen);
+    float y = y_gen(gen);
+
+    float dest_x = BOARD_LENGTH + 30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    float x = x_gen_right(gen);
+    float y = y_gen(gen);
+
+    float dest_x = -30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int j = n_p; j < n_p_pad; j++) {
+    pos_x[j] = 1e9f;
+    pos_y[j] = 1e9f;
+    sb_sq[j] = 0.0f;
+    sb_e_x[j] = 0.0f;
+    sb_e_y[j] = 0.0f;
+  }
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+// # 256 MULTs, 224 ADDs
+static inline __m512 _avx512_exp_ps(__m512 x) {
+
+  const __m512 log2e = _mm512_set1_ps(1.4426950408889634f);
+
+  // # 16 MULTs
+  __m512 nf = _mm512_roundscale_ps(_mm512_mul_ps(x, log2e), _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+
+  const __m512 ln2 = _mm512_set1_ps(0.6931471805599453f);
+
+  // # 16 MULTs, 16 ADDs
+  __m512 r = _mm512_fnmadd_ps(nf, ln2, x);
+
+  const __m512 c13 = _mm512_set1_ps(c13_scalar);
+  const __m512 c12 = _mm512_set1_ps(c12_scalar);
+  const __m512 c11 = _mm512_set1_ps(c11_scalar);
+  const __m512 c10 = _mm512_set1_ps(c10_scalar);
+  const __m512 c9 = _mm512_set1_ps(c9_scalar);
+  const __m512 c8 = _mm512_set1_ps(c8_scalar);
+  const __m512 c7 = _mm512_set1_ps(c7_scalar);
+  const __m512 c6 = _mm512_set1_ps(c6_scalar);
+  const __m512 c5 = _mm512_set1_ps(c5_scalar);
+  const __m512 c4 = _mm512_set1_ps(c4_scalar);
+  const __m512 c3 = _mm512_set1_ps(c3_scalar);
+  const __m512 c2 = _mm512_set1_ps(c2_scalar);
+  const __m512 c1 = _mm512_set1_ps(c1_scalar);
+  const __m512 c0 = _mm512_set1_ps(c0_scalar);
+
+  __m512 p = c13;
+
+  // #   13 * (16 MULTs, 16 ADDs)
+  // # = 208 MULTs, 208 ADDs
+  p = _mm512_fmadd_ps(p, r, c12);
+  p = _mm512_fmadd_ps(p, r, c11);
+  p = _mm512_fmadd_ps(p, r, c10);
+  p = _mm512_fmadd_ps(p, r, c9);
+  p = _mm512_fmadd_ps(p, r, c8);
+  p = _mm512_fmadd_ps(p, r, c7);
+  p = _mm512_fmadd_ps(p, r, c6);
+  p = _mm512_fmadd_ps(p, r, c5);
+  p = _mm512_fmadd_ps(p, r, c4);
+  p = _mm512_fmadd_ps(p, r, c3);
+  p = _mm512_fmadd_ps(p, r, c2);
+  p = _mm512_fmadd_ps(p, r, c1);
+  p = _mm512_fmadd_ps(p, r, c0);
+
+  __m512i ni = _mm512_cvtps_epi32(nf);
+  __m512i biased = _mm512_add_epi32(ni, _mm512_set1_epi32(127));
+  __m512i scaled = _mm512_slli_epi32(biased, 23);
+  __m512 pow2n = _mm512_castsi512_ps(scaled);
+
+  // # 16 MULTs
+  return _mm512_mul_ps(p, pow2n);
+}
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+  (void)pedestrians;
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+  vector<tuple<int, int, float, float, float, float, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+    for (int i = 0; i < n_p; i++) {
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  const int n_p_pad = (n_p + 31) & ~31;
+
+  alignas(64) const float lane_offset_arr[16] = {0.0f, 1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f};
+  const __m512 lane_off = _mm512_load_ps(lane_offset_arr);
+
+  const __m512 v_neg_v0_div_2s = _mm512_set1_ps(NEG_V0_AB_div_TWO_SIGMA);
+  const __m512 v_neg_2s_inv = _mm512_set1_ps(NEG_TWO_SIGMA_INV);
+  const __m512 v_cos_phi_sq = _mm512_set1_ps(cos_phi_sq);
+  const __m512 v_dt_neg = _mm512_set1_ps(dt_neg);
+  const __m512 v_c_neg_dt = _mm512_set1_ps(c_neg_times_dt);
+  const __m512 v_zero = _mm512_setzero_ps();
+  const __m512 v_one = _mm512_set1_ps(1.0f);
+  const __m512 v_eps = _mm512_set1_ps(EPS);
+  const __m512 v_eps_sq = _mm512_set1_ps(EPS_SQ);
+  const __m512 v_np = _mm512_set1_ps((float)n_p);
+  const __m512 v_exp_min = _mm512_set1_ps(-87.0f);
+
+  // # ADDs: n_iter * (
+  // #   n_p * 68 +
+  // #   n_p * n_p_pad * 29 +
+  // #   n_p * board.n_obstacles * 10 +
+  // #   n_p_pad * 7
+  // # )
+
+  // # MULTs: n_iter * (
+  // #   n_p * 4 +
+  // #   n_p * n_p_pad * 40 +
+  // #   n_p * board.n_obstacles * 13 +
+  // #   n_p_pad * 15
+  // # )
+
+  // # DIVs: n_iter * (
+  // #   n_p * n_p_pad +
+  // #   n_p * board.n_obstacles +
+  // #   n_p_pad * 2
+  // # )
+
+  // # SQRTs:  n_iter * (
+  // #   n_p * n_p_pad * 3 +
+  // #   n_p * board.n_obstacles +
+  // #   n_p_pad * 3
+  // # )
+
+  // # EXPs: n_iter * n_p * board.n_obstacles
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+
+      // # 2 MULTs
+      float desired_velocity_x = e_x[i] * desired_v0[i];
+      float desired_velocity_y = e_y[i] * desired_v0[i];
+
+      // # 2 ADDs, 2 MULTs
+      float f_alpha_dt_x = (desired_velocity_x - curr_vel_x[i]) * dt_times_tau_alpha_inv;
+      float f_alpha_dt_y = (desired_velocity_y - curr_vel_y[i]) * dt_times_tau_alpha_inv;
+
+      const __m512 pix = _mm512_set1_ps(pos_x[i]);
+      const __m512 piy = _mm512_set1_ps(pos_y[i]);
+      const __m512 eix = _mm512_set1_ps(e_x[i]);
+      const __m512 eiy = _mm512_set1_ps(e_y[i]);
+      const __m512 v_i = _mm512_set1_ps((float)i);
+
+      __m512 acc_x_0 = _mm512_setzero_ps();
+      __m512 acc_y_0 = _mm512_setzero_ps();
+      __m512 acc_x_1 = _mm512_setzero_ps();
+      __m512 acc_y_1 = _mm512_setzero_ps();
+
+      for (int j = 0; j < n_p_pad; j += 32) {
+
+        // # 32 ADDs
+        __m512 j_idx_0 = _mm512_add_ps(_mm512_set1_ps((float)j), lane_off);
+        __m512 j_idx_1 = _mm512_add_ps(_mm512_set1_ps((float)j + 16), lane_off);
+
+        __mmask16 m_neq_i_0 = _mm512_cmp_ps_mask(j_idx_0, v_i, _CMP_NEQ_OQ);
+        __mmask16 m_neq_i_1 = _mm512_cmp_ps_mask(j_idx_1, v_i, _CMP_NEQ_OQ);
+
+        __mmask16 m_in_range_0 = _mm512_cmp_ps_mask(j_idx_0, v_np, _CMP_LT_OQ);
+        __mmask16 m_in_range_1 = _mm512_cmp_ps_mask(j_idx_1, v_np, _CMP_LT_OQ);
+
+        __mmask16 active_0 = m_neq_i_0 & m_in_range_0;
+        __mmask16 active_1 = m_neq_i_1 & m_in_range_1;
+
+        __m512 pjx_0 = _mm512_loadu_ps(&pos_x[j]);
+        __m512 pjx_1 = _mm512_loadu_ps(&pos_x[j + 16]);
+
+        __m512 pjy_0 = _mm512_loadu_ps(&pos_y[j]);
+        __m512 pjy_1 = _mm512_loadu_ps(&pos_y[j + 16]);
+
+        __m512 sbe_jx_0 = _mm512_loadu_ps(&sb_e_x[j]);
+        __m512 sbe_jx_1 = _mm512_loadu_ps(&sb_e_x[j + 16]);
+
+        __m512 sbe_jy_0 = _mm512_loadu_ps(&sb_e_y[j]);
+        __m512 sbe_jy_1 = _mm512_loadu_ps(&sb_e_y[j + 16]);
+
+        __m512 sb_jsq_0 = _mm512_loadu_ps(&sb_sq[j]);
+        __m512 sb_jsq_1 = _mm512_loadu_ps(&sb_sq[j + 16]);
+
+        // # 32 ADDs
+        __m512 r_ab_x_0 = _mm512_sub_ps(pix, pjx_0);
+        __m512 r_ab_x_1 = _mm512_sub_ps(pix, pjx_1);
+
+        // # 32 ADDs
+        __m512 r_ab_y_0 = _mm512_sub_ps(piy, pjy_0);
+        __m512 r_ab_y_1 = _mm512_sub_ps(piy, pjy_1);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 r_ab_n2_0 = _mm512_fmadd_ps(r_ab_x_0, r_ab_x_0, _mm512_mul_ps(r_ab_y_0, r_ab_y_0));
+        __m512 r_ab_n2_1 = _mm512_fmadd_ps(r_ab_x_1, r_ab_x_1, _mm512_mul_ps(r_ab_y_1, r_ab_y_1));
+
+        // # 32 SQRTs
+        __m512 r_ab_n_0 = _mm512_sqrt_ps(r_ab_n2_0);
+        __m512 r_ab_n_1 = _mm512_sqrt_ps(r_ab_n2_1);
+
+        // # 32 ADDs
+        __m512 r_ab2_x_0 = _mm512_sub_ps(r_ab_x_0, sbe_jx_0);
+        __m512 r_ab2_x_1 = _mm512_sub_ps(r_ab_x_1, sbe_jx_1);
+
+        // # 32 ADDs
+        __m512 r_ab2_y_0 = _mm512_sub_ps(r_ab_y_0, sbe_jy_0);
+        __m512 r_ab2_y_1 = _mm512_sub_ps(r_ab_y_1, sbe_jy_1);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 r_ab2_n2_0 = _mm512_fmadd_ps(r_ab2_x_0, r_ab2_x_0, _mm512_mul_ps(r_ab2_y_0, r_ab2_y_0));
+        __m512 r_ab2_n2_1 = _mm512_fmadd_ps(r_ab2_x_1, r_ab2_x_1, _mm512_mul_ps(r_ab2_y_1, r_ab2_y_1));
+
+        // # 32 SQRTs
+        __m512 r_ab2_n_0 = _mm512_sqrt_ps(r_ab2_n2_0);
+        __m512 r_ab2_n_1 = _mm512_sqrt_ps(r_ab2_n2_1);
+
+        // # 32 ADDs
+        __m512 r_sum_0 = _mm512_add_ps(r_ab_n_0, r_ab2_n_0);
+        __m512 r_sum_1 = _mm512_add_ps(r_ab_n_1, r_ab2_n_1);
+
+        // # 32 MULTs, 32 ADDs
+        __m512 two_b_sq_0 = _mm512_fmsub_ps(r_sum_0, r_sum_0, sb_jsq_0);
+        __m512 two_b_sq_1 = _mm512_fmsub_ps(r_sum_1, r_sum_1, sb_jsq_1);
+
+        // # 32 SQRTs
+        __m512 two_b_0 = _mm512_sqrt_ps(two_b_sq_0);
+        __m512 two_b_1 = _mm512_sqrt_ps(two_b_sq_1);
+
+        // # 32 MULTs
+        __m512 e_arg_0 = _mm512_mul_ps(two_b_0, v_neg_2s_inv);
+        __m512 e_arg_1 = _mm512_mul_ps(two_b_1, v_neg_2s_inv);
+
+        e_arg_0 = _mm512_max_ps(e_arg_0, v_exp_min);
+        e_arg_1 = _mm512_max_ps(e_arg_1, v_exp_min);
+
+        // # 512 MULTs, 448 ADDs
+        __m512 e_val_0 = _avx512_exp_ps(e_arg_0);
+        __m512 e_val_1 = _avx512_exp_ps(e_arg_1);
+
+        // # 64 MULTs
+        __m512 numer_0 = _mm512_mul_ps(_mm512_mul_ps(r_sum_0, v_neg_v0_div_2s), e_val_0);
+        __m512 numer_1 = _mm512_mul_ps(_mm512_mul_ps(r_sum_1, v_neg_v0_div_2s), e_val_1);
+
+        // # 64 MULTs
+        __m512 denom_0 = _mm512_mul_ps(_mm512_mul_ps(two_b_0, r_ab_n_0), r_ab2_n_0);
+        __m512 denom_1 = _mm512_mul_ps(_mm512_mul_ps(two_b_1, r_ab_n_1), r_ab2_n_1);
+
+        __m512 denom_safe_0 = _mm512_mask_blend_ps(active_0, v_one, denom_0);
+        __m512 denom_safe_1 = _mm512_mask_blend_ps(active_1, v_one, denom_1);
+
+        // # 32 DIVs
+        __m512 smc_0 = _mm512_div_ps(numer_0, denom_safe_0);
+        __m512 smc_1 = _mm512_div_ps(numer_1, denom_safe_1);
+
+        // # 32 MULTs
+        __m512 sm_r_ab_0 = _mm512_mul_ps(smc_0, r_ab2_n_0);
+        __m512 sm_r_ab_1 = _mm512_mul_ps(smc_1, r_ab2_n_1);
+
+        // # 32 MULTs
+        __m512 sm_r_ab2_0 = _mm512_mul_ps(smc_0, r_ab_n_0);
+        __m512 sm_r_ab2_1 = _mm512_mul_ps(smc_1, r_ab_n_1);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 f_x_0 = _mm512_fmadd_ps(r_ab2_x_0, sm_r_ab2_0, _mm512_mul_ps(r_ab_x_0, sm_r_ab_0));
+        __m512 f_x_1 = _mm512_fmadd_ps(r_ab2_x_1, sm_r_ab2_1, _mm512_mul_ps(r_ab_x_1, sm_r_ab_1));
+
+        // # 64 MULTs, 32 ADDs
+        __m512 f_y_0 = _mm512_fmadd_ps(r_ab2_y_0, sm_r_ab2_0, _mm512_mul_ps(r_ab_y_0, sm_r_ab_0));
+        __m512 f_y_1 = _mm512_fmadd_ps(r_ab2_y_1, sm_r_ab2_1, _mm512_mul_ps(r_ab_y_1, sm_r_ab_1));
+
+        // # 64 MULTs, 32 ADDs
+        __m512 f_norm_sq_0 = _mm512_fmadd_ps(f_x_0, f_x_0, _mm512_mul_ps(f_y_0, f_y_0));
+        __m512 f_norm_sq_1 = _mm512_fmadd_ps(f_x_1, f_x_1, _mm512_mul_ps(f_y_1, f_y_1));
+
+        // # 32 MULTs
+        __m512 rhs_0 = _mm512_mul_ps(f_norm_sq_0, v_cos_phi_sq);
+        __m512 rhs_1 = _mm512_mul_ps(f_norm_sq_1, v_cos_phi_sq);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 e_dot_f_0 = _mm512_fmadd_ps(eix, f_x_0, _mm512_mul_ps(eiy, f_y_0));
+        __m512 e_dot_f_1 = _mm512_fmadd_ps(eix, f_x_1, _mm512_mul_ps(eiy, f_y_1));
+
+        // # 32 MULTs
+        __m512 e_dot_f_sq_0 = _mm512_mul_ps(e_dot_f_0, e_dot_f_0);
+        __m512 e_dot_f_sq_1 = _mm512_mul_ps(e_dot_f_1, e_dot_f_1);
+
+        __mmask16 c1_0 = _mm512_cmp_ps_mask(e_dot_f_0, v_zero, _CMP_LT_OQ);
+        __mmask16 c1_1 = _mm512_cmp_ps_mask(e_dot_f_1, v_zero, _CMP_LT_OQ);
+
+        __mmask16 c2_0 = _mm512_cmp_ps_mask(e_dot_f_sq_0, rhs_0, _CMP_GT_OQ);
+        __mmask16 c2_1 = _mm512_cmp_ps_mask(e_dot_f_sq_1, rhs_1, _CMP_GT_OQ);
+
+        __mmask16 cond_0 = c1_0 & c2_0;
+        __mmask16 cond_1 = c1_1 & c2_1;
+
+        __m512 w_val_0 = _mm512_mask_blend_ps(cond_0, v_dt_neg, v_c_neg_dt);
+        __m512 w_val_1 = _mm512_mask_blend_ps(cond_1, v_dt_neg, v_c_neg_dt);
+
+        __m512 f_x_safe_0 = _mm512_maskz_mov_ps(active_0, f_x_0);
+        __m512 f_x_safe_1 = _mm512_maskz_mov_ps(active_1, f_x_1);
+
+        __m512 f_y_safe_0 = _mm512_maskz_mov_ps(active_0, f_y_0);
+        __m512 f_y_safe_1 = _mm512_maskz_mov_ps(active_1, f_y_1);
+
+        // # 32 MULTs, 32 ADDs
+        acc_x_0 = _mm512_fmadd_ps(f_x_safe_0, w_val_0, acc_x_0);
+        acc_x_1 = _mm512_fmadd_ps(f_x_safe_1, w_val_1, acc_x_1);
+
+        // # 32 MULTs, 32 ADDs
+        acc_y_0 = _mm512_fmadd_ps(f_y_safe_0, w_val_0, acc_y_0);
+        acc_y_1 = _mm512_fmadd_ps(f_y_safe_1, w_val_1, acc_y_1);
+      }
+
+      // # 64 ADDs
+      f_alpha_dt_x += _mm512_reduce_add_ps(_mm512_add_ps(acc_x_0, acc_x_1));
+      f_alpha_dt_y += _mm512_reduce_add_ps(_mm512_add_ps(acc_y_0, acc_y_1));
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
+        float v1_x = pos_x[i] - board.obstacles[k].start_x;
+        float v1_y = pos_y[i] - board.obstacles[k].start_y;
+        float v2_x = board.obstacles[k].obstacle_v_x;
+        float v2_y = board.obstacles[k].obstacle_v_y;
+
+        // # 3 MULTs, 1 ADD
+        float projection = (v1_x * v2_x + v1_y * v2_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0f)
+          projection = 0.0f;
+        else if (projection > 1.0f)
+          projection = 1.0f;
+
+        // # 4 ADDs, 2 MULTs
+        float r_aB_x = pos_x[i] - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        float r_aB_y = pos_y[i] - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        float r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+        r_aB_norm = r_aB_norm > EPS ? r_aB_norm : EPS;
+
+        // # 4 MULTs, 1 EXP, 1 DIV
+        float scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+        float f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        float f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        float w_val = dt_neg;
+        float e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -e_y[i] : e_y[i]);
+
+        if (e_dot_f < cos_phi)
+          w_val = c_neg_times_dt;
+
+        // # 2 MULTS
+        float f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        float f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        // # 2 ADDs
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      // # 2 ADDs
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p_pad; i += 32) {
+
+      __m512 w_x_i_vec_0 = _mm512_loadu_ps(&w_x[i]);
+      __m512 w_x_i_vec_1 = _mm512_loadu_ps(&w_x[i + 16]);
+
+      __m512 w_y_i_vec_0 = _mm512_loadu_ps(&w_y[i]);
+      __m512 w_y_i_vec_1 = _mm512_loadu_ps(&w_y[i + 16]);
+
+      __m512 v_max_i_vec_0 = _mm512_loadu_ps(&v_max[i]);
+      __m512 v_max_i_vec_1 = _mm512_loadu_ps(&v_max[i + 16]);
+
+      __m512 pos_x_i_vec_0 = _mm512_loadu_ps(&pos_x[i]);
+      __m512 pos_x_i_vec_1 = _mm512_loadu_ps(&pos_x[i + 16]);
+
+      __m512 pos_y_i_vec_0 = _mm512_loadu_ps(&pos_y[i]);
+      __m512 pos_y_i_vec_1 = _mm512_loadu_ps(&pos_y[i + 16]);
+
+      __m512 destination_y_i_vec_0 = _mm512_loadu_ps(&destination_y[i]);
+      __m512 destination_y_i_vec_1 = _mm512_loadu_ps(&destination_y[i + 16]);
+
+      __m512 destination_x_i_vec_0 = _mm512_loadu_ps(&destination_x[i]);
+      __m512 destination_x_i_vec_1 = _mm512_loadu_ps(&destination_x[i + 16]);
+
+      __m512 desired_v0_i_vec_0 = _mm512_loadu_ps(&desired_v0[i]);
+      __m512 desired_v0_i_vec_1 = _mm512_loadu_ps(&desired_v0[i + 16]);
+
+      __m512i color_i_vec_0 = _mm512_loadu_si512((__m512i *)&color[i]);
+      __m512i color_i_vec_1 = _mm512_loadu_si512((__m512i *)&color[i + 16]);
+
+      // # 64 MULTs, 32 ADDs
+      __m512 w_norm_sq_0 = _mm512_fmadd_ps(w_x_i_vec_0, w_x_i_vec_0, _mm512_mul_ps(w_y_i_vec_0, w_y_i_vec_0));
+      __m512 w_norm_sq_1 = _mm512_fmadd_ps(w_x_i_vec_1, w_x_i_vec_1, _mm512_mul_ps(w_y_i_vec_1, w_y_i_vec_1));
+
+      // # 32 SQRTs
+      __m512 w_norm_0 = _mm512_sqrt_ps(w_norm_sq_0);
+      __m512 w_norm_1 = _mm512_sqrt_ps(w_norm_sq_1);
+
+      // # 32 DIVs
+      __m512 correction_factor_0 = _mm512_min_ps(_mm512_set1_ps(1.0f), _mm512_div_ps(v_max_i_vec_0, w_norm_0));
+      __m512 correction_factor_1 = _mm512_min_ps(_mm512_set1_ps(1.0f), _mm512_div_ps(v_max_i_vec_1, w_norm_1));
+
+      // # 32 MULTs
+      __m512 curr_vel_x_i_vec_0 = _mm512_mul_ps(w_x_i_vec_0, correction_factor_0);
+      __m512 curr_vel_x_i_vec_1 = _mm512_mul_ps(w_x_i_vec_1, correction_factor_1);
+
+      // # 32 MULTs
+      __m512 curr_vel_y_i_vec_0 = _mm512_mul_ps(w_y_i_vec_0, correction_factor_0);
+      __m512 curr_vel_y_i_vec_1 = _mm512_mul_ps(w_y_i_vec_1, correction_factor_1);
+
+      __m512 dt_vec = _mm512_set1_ps(dt);
+
+      // # 32 MULTs, 32 ADDs
+      pos_x_i_vec_0 = _mm512_fmadd_ps(curr_vel_x_i_vec_0, dt_vec, pos_x_i_vec_0);
+      pos_x_i_vec_1 = _mm512_fmadd_ps(curr_vel_x_i_vec_1, dt_vec, pos_x_i_vec_1);
+
+      // # 32 MULTs, 32 ADDs
+      pos_y_i_vec_0 = _mm512_fmadd_ps(curr_vel_y_i_vec_0, dt_vec, pos_y_i_vec_0);
+      pos_y_i_vec_1 = _mm512_fmadd_ps(curr_vel_y_i_vec_1, dt_vec, pos_y_i_vec_1);
+
+      __mmask16 cond_1_left_0 = _mm512_cmpeq_epi32_mask(color_i_vec_0, _mm512_setzero_si512());
+      __mmask16 cond_1_left_1 = _mm512_cmpeq_epi32_mask(color_i_vec_1, _mm512_setzero_si512());
+
+      __mmask16 cond_1_right_0 = _mm512_cmp_ps_mask(pos_x_i_vec_0, _mm512_set1_ps(BOARD_LENGTH_MINUS_ONE_POINT_FIVE), _CMP_GT_OQ);
+      __mmask16 cond_1_right_1 = _mm512_cmp_ps_mask(pos_x_i_vec_1, _mm512_set1_ps(BOARD_LENGTH_MINUS_ONE_POINT_FIVE), _CMP_GT_OQ);
+
+      __mmask16 cond_1_0 = cond_1_left_0 & cond_1_right_0;
+      __mmask16 cond_1_1 = cond_1_left_1 & cond_1_right_1;
+
+      pos_x_i_vec_0 = _mm512_mask_blend_ps(cond_1_0, pos_x_i_vec_0, _mm512_set1_ps(1.0f));
+      pos_x_i_vec_1 = _mm512_mask_blend_ps(cond_1_1, pos_x_i_vec_1, _mm512_set1_ps(1.0f));
+
+      destination_y_i_vec_0 = _mm512_mask_blend_ps(cond_1_0, destination_y_i_vec_0, pos_y_i_vec_0);
+      destination_y_i_vec_1 = _mm512_mask_blend_ps(cond_1_1, destination_y_i_vec_1, pos_y_i_vec_1);
+
+      w_x_i_vec_0 = _mm512_mask_blend_ps(cond_1_0, w_x_i_vec_0, desired_v0_i_vec_0);
+      w_x_i_vec_1 = _mm512_mask_blend_ps(cond_1_1, w_x_i_vec_1, desired_v0_i_vec_1);
+
+      w_y_i_vec_0 = _mm512_mask_blend_ps(cond_1_0, w_y_i_vec_0, _mm512_setzero_ps());
+      w_y_i_vec_1 = _mm512_mask_blend_ps(cond_1_1, w_y_i_vec_1, _mm512_setzero_ps());
+
+      curr_vel_x_i_vec_0 = _mm512_mask_blend_ps(cond_1_0, curr_vel_x_i_vec_0, w_x_i_vec_0);
+      curr_vel_x_i_vec_1 = _mm512_mask_blend_ps(cond_1_1, curr_vel_x_i_vec_1, w_x_i_vec_1);
+
+      curr_vel_y_i_vec_0 = _mm512_mask_blend_ps(cond_1_0, curr_vel_y_i_vec_0, w_y_i_vec_0);
+      curr_vel_y_i_vec_1 = _mm512_mask_blend_ps(cond_1_1, curr_vel_y_i_vec_1, w_y_i_vec_1);
+
+      __mmask16 cond_2_left_0 = _mm512_cmpeq_epi32_mask(color_i_vec_0, _mm512_set1_epi32(1));
+      __mmask16 cond_2_left_1 = _mm512_cmpeq_epi32_mask(color_i_vec_1, _mm512_set1_epi32(1));
+
+      __mmask16 cond_2_right_0 = _mm512_cmp_ps_mask(pos_x_i_vec_0, _mm512_set1_ps(1.5f), _CMP_LT_OQ);
+      __mmask16 cond_2_right_1 = _mm512_cmp_ps_mask(pos_x_i_vec_1, _mm512_set1_ps(1.5f), _CMP_LT_OQ);
+
+      __mmask16 cond_2_0 = cond_2_left_0 & cond_2_right_0;
+      __mmask16 cond_2_1 = cond_2_left_1 & cond_2_right_1;
+
+      pos_x_i_vec_0 = _mm512_mask_blend_ps(cond_2_0, pos_x_i_vec_0, _mm512_set1_ps(BOARD_LENGTH - 1.0f));
+      pos_x_i_vec_1 = _mm512_mask_blend_ps(cond_2_1, pos_x_i_vec_1, _mm512_set1_ps(BOARD_LENGTH - 1.0f));
+
+      destination_y_i_vec_0 = _mm512_mask_blend_ps(cond_2_0, destination_y_i_vec_0, pos_y_i_vec_0);
+      destination_y_i_vec_1 = _mm512_mask_blend_ps(cond_2_1, destination_y_i_vec_1, pos_y_i_vec_1);
+
+      __m512 neg_desired_v0_0 = _mm512_xor_ps(desired_v0_i_vec_0, _mm512_set1_ps(-0.0f));
+      __m512 neg_desired_v0_1 = _mm512_xor_ps(desired_v0_i_vec_1, _mm512_set1_ps(-0.0f));
+
+      w_x_i_vec_0 = _mm512_mask_blend_ps(cond_2_0, w_x_i_vec_0, neg_desired_v0_0);
+      w_x_i_vec_1 = _mm512_mask_blend_ps(cond_2_1, w_x_i_vec_1, neg_desired_v0_1);
+
+      w_y_i_vec_0 = _mm512_mask_blend_ps(cond_2_0, w_y_i_vec_0, _mm512_setzero_ps());
+      w_y_i_vec_1 = _mm512_mask_blend_ps(cond_2_1, w_y_i_vec_1, _mm512_setzero_ps());
+
+      curr_vel_x_i_vec_0 = _mm512_mask_blend_ps(cond_2_0, curr_vel_x_i_vec_0, w_x_i_vec_0);
+      curr_vel_y_i_vec_1 = _mm512_mask_blend_ps(cond_2_1, curr_vel_y_i_vec_1, w_y_i_vec_1);
+
+      pos_y_i_vec_0 = _mm512_max_ps(pos_y_i_vec_0, _mm512_set1_ps(WALL_EPS));
+      pos_y_i_vec_1 = _mm512_max_ps(pos_y_i_vec_1, _mm512_set1_ps(WALL_EPS));
+
+      pos_y_i_vec_0 = _mm512_min_ps(pos_y_i_vec_0, _mm512_set1_ps(BOARD_WIDTH_MINUS_WALL_EPS));
+      pos_y_i_vec_1 = _mm512_min_ps(pos_y_i_vec_1, _mm512_set1_ps(BOARD_WIDTH_MINUS_WALL_EPS));
+
+      // # 32 ADDs
+      __m512 ex_vec_0 = _mm512_sub_ps(destination_x_i_vec_0, pos_x_i_vec_0);
+      __m512 ex_vec_1 = _mm512_sub_ps(destination_x_i_vec_1, pos_x_i_vec_1);
+
+      // # 32 ADDs
+      __m512 ey_vec_0 = _mm512_sub_ps(destination_y_i_vec_0, pos_y_i_vec_0);
+      __m512 ey_vec_1 = _mm512_sub_ps(destination_y_i_vec_1, pos_y_i_vec_1);
+
+      // # 64 MULTs, 32 ADDs
+      __m512 e_norm_sq_0 = _mm512_fmadd_ps(ex_vec_0, ex_vec_0, _mm512_mul_ps(ey_vec_0, ey_vec_0));
+      __m512 e_norm_sq_1 = _mm512_fmadd_ps(ex_vec_1, ex_vec_1, _mm512_mul_ps(ey_vec_1, ey_vec_1));
+
+      // # 32 SQRTs
+      __m512 e_norm_0 = _mm512_sqrt_ps(e_norm_sq_0);
+      __m512 e_norm_1 = _mm512_sqrt_ps(e_norm_sq_1);
+
+      // # 32 DIVs
+      __m512 e_norm_inv_0 = _mm512_div_ps(_mm512_set1_ps(1.0f), e_norm_0);
+      __m512 e_norm_inv_1 = _mm512_div_ps(_mm512_set1_ps(1.0f), e_norm_1);
+
+      // # 32 MULTs
+      __m512 e_x_i_vec_0 = _mm512_mul_ps(ex_vec_0, e_norm_inv_0);
+      __m512 e_x_i_vec_1 = _mm512_mul_ps(ex_vec_1, e_norm_inv_1);
+
+      // # 32 MULTs
+      __m512 e_y_i_vec_0 = _mm512_mul_ps(ey_vec_0, e_norm_inv_0);
+      __m512 e_y_i_vec_1 = _mm512_mul_ps(ey_vec_1, e_norm_inv_1);
+
+      // # 64 MULTs, 32 ADDs
+      __m512 vel_norm_sq_0 = _mm512_fmadd_ps(curr_vel_x_i_vec_0, curr_vel_x_i_vec_0, _mm512_mul_ps(curr_vel_y_i_vec_0, curr_vel_y_i_vec_0));
+      __m512 vel_norm_sq_1 = _mm512_fmadd_ps(curr_vel_x_i_vec_1, curr_vel_x_i_vec_1, _mm512_mul_ps(curr_vel_y_i_vec_1, curr_vel_y_i_vec_1));
+
+      // # 32 MULTs
+      __m512 sb_sq_i_vec_0 = _mm512_mul_ps(vel_norm_sq_0, _mm512_set1_ps(delta_t_sq));
+      __m512 sb_sq_i_vec_1 = _mm512_mul_ps(vel_norm_sq_1, _mm512_set1_ps(delta_t_sq));
+
+      // # 32 SQRTs
+      __m512 vel_norm_dt_0 = _mm512_sqrt_ps(sb_sq_i_vec_0);
+      __m512 vel_norm_dt_1 = _mm512_sqrt_ps(sb_sq_i_vec_1);
+
+      _mm512_storeu_ps(&curr_vel_x[i], curr_vel_x_i_vec_0);
+      _mm512_storeu_ps(&curr_vel_x[i + 16], curr_vel_x_i_vec_1);
+
+      _mm512_storeu_ps(&curr_vel_y[i], curr_vel_y_i_vec_0);
+      _mm512_storeu_ps(&curr_vel_y[i + 16], curr_vel_y_i_vec_1);
+
+      _mm512_storeu_ps(&w_x[i], w_x_i_vec_0);
+      _mm512_storeu_ps(&w_x[i + 16], w_x_i_vec_1);
+
+      _mm512_storeu_ps(&w_y[i], w_y_i_vec_0);
+      _mm512_storeu_ps(&w_y[i + 16], w_y_i_vec_1);
+
+      _mm512_storeu_ps(&pos_x[i], pos_x_i_vec_0);
+      _mm512_storeu_ps(&pos_x[i + 16], pos_x_i_vec_1);
+
+      _mm512_storeu_ps(&pos_y[i], pos_y_i_vec_0);
+      _mm512_storeu_ps(&pos_y[i + 16], pos_y_i_vec_1);
+
+      _mm512_storeu_ps(&e_x[i], e_x_i_vec_0);
+      _mm512_storeu_ps(&e_x[i + 16], e_x_i_vec_1);
+
+      _mm512_storeu_ps(&e_y[i], e_y_i_vec_0);
+      _mm512_storeu_ps(&e_y[i + 16], e_y_i_vec_1);
+
+      _mm512_storeu_ps(&destination_y[i], destination_y_i_vec_0);
+      _mm512_storeu_ps(&destination_y[i + 16], destination_y_i_vec_1);
+
+      _mm512_storeu_ps(&sb_sq[i], sb_sq_i_vec_0);
+      _mm512_storeu_ps(&sb_sq[i + 16], sb_sq_i_vec_1);
+
+      // # 32 MULTs
+      _mm512_storeu_ps(&sb_e_x[i], _mm512_mul_ps(e_x_i_vec_0, vel_norm_dt_0));
+      _mm512_storeu_ps(&sb_e_x[i + 16], _mm512_mul_ps(e_x_i_vec_1, vel_norm_dt_1));
+
+      // # 32 MULTs
+      _mm512_storeu_ps(&sb_e_y[i], _mm512_mul_ps(e_y_i_vec_0, vel_norm_dt_0));
+      _mm512_storeu_ps(&sb_e_y[i + 16], _mm512_mul_ps(e_y_i_vec_1, vel_norm_dt_1));
+    }
+
+#if !BENCHMARK
+    for (int i = 0; i < n_p; i++) {
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+    }
+#endif
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix.cpp b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix.cpp
new file mode 100644
index 0000000..ec69800
--- /dev/null
+++ b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_expfix.cpp
@@ -0,0 +1,634 @@
+// # Changes: Used _mm512_rsqrt14_ps() instead of _mm512_sqrt_ps() and _mm512_div_ps()
+// # Predecessor: variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2.cpp
+
+#include "../headers/board_reduced_float.hpp"
+#include "../headers/constants_float.hpp"
+#include "../headers/pedestrian_reduced_float.hpp"
+
+#include <immintrin.h>
+
+#include <fstream>
+#include <random>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0f
+#define BOARD_LENGTH 50.0f
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0f
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5f
+#define WALL_EPS 1e-3f
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999f
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+alignas(64) vector<float> pos_x, pos_y;
+alignas(64) vector<float> sb_sq;
+alignas(64) vector<float> sb_e_x, sb_e_y;
+alignas(64) vector<float> e_x, e_y;
+alignas(64) vector<float> curr_vel_x, curr_vel_y;
+alignas(64) vector<float> w_x, w_y;
+alignas(64) vector<float> desired_v0;
+alignas(64) vector<float> v_max;
+alignas(64) vector<float> destination_x, destination_y;
+alignas(64) vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<float> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<float> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<float> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<float> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  const int n_p_pad = (n_p + 31) & ~31;
+
+  pos_x.resize(n_p_pad);
+  pos_y.resize(n_p_pad);
+  sb_sq.resize(n_p_pad);
+  sb_e_x.resize(n_p_pad);
+  sb_e_y.resize(n_p_pad);
+  e_x.resize(n_p_pad);
+  e_y.resize(n_p_pad);
+  curr_vel_x.resize(n_p_pad);
+  curr_vel_y.resize(n_p_pad);
+  w_x.resize(n_p_pad);
+  w_y.resize(n_p_pad);
+  desired_v0.resize(n_p_pad);
+  v_max.resize(n_p_pad);
+  destination_x.resize(n_p_pad);
+  destination_y.resize(n_p_pad);
+  color.resize(n_p_pad);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    float x = x_gen_left(gen);
+    float y = y_gen(gen);
+
+    float dest_x = BOARD_LENGTH + 30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    float x = x_gen_right(gen);
+    float y = y_gen(gen);
+
+    float dest_x = -30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int j = n_p; j < n_p_pad; j++) {
+    pos_x[j] = 1e9f;
+    pos_y[j] = 1e9f;
+    sb_sq[j] = 0.0f;
+    sb_e_x[j] = 0.0f;
+    sb_e_y[j] = 0.0f;
+  }
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+// # 256 MULTs, 224 ADDs
+static inline __m512 _avx512_exp_ps(__m512 x) {
+
+  const __m512 log2e = _mm512_set1_ps(1.4426950408889634f);
+
+  // # 16 MULTs
+  __m512 nf = _mm512_roundscale_ps(_mm512_mul_ps(x, log2e), _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+
+  const __m512 ln2 = _mm512_set1_ps(0.6931471805599453f);
+
+  // # 16 MULTs, 16 ADDs
+  __m512 r = _mm512_fnmadd_ps(nf, ln2, x);
+
+  const __m512 c13 = _mm512_set1_ps(c13_scalar);
+  const __m512 c12 = _mm512_set1_ps(c12_scalar);
+  const __m512 c11 = _mm512_set1_ps(c11_scalar);
+  const __m512 c10 = _mm512_set1_ps(c10_scalar);
+  const __m512 c9 = _mm512_set1_ps(c9_scalar);
+  const __m512 c8 = _mm512_set1_ps(c8_scalar);
+  const __m512 c7 = _mm512_set1_ps(c7_scalar);
+  const __m512 c6 = _mm512_set1_ps(c6_scalar);
+  const __m512 c5 = _mm512_set1_ps(c5_scalar);
+  const __m512 c4 = _mm512_set1_ps(c4_scalar);
+  const __m512 c3 = _mm512_set1_ps(c3_scalar);
+  const __m512 c2 = _mm512_set1_ps(c2_scalar);
+  const __m512 c1 = _mm512_set1_ps(c1_scalar);
+  const __m512 c0 = _mm512_set1_ps(c0_scalar);
+
+  __m512 p = c13;
+
+  // #   13 * (16 MULTs, 16 ADDs)
+  // # = 208 MULTs, 208 ADDs
+  p = _mm512_fmadd_ps(p, r, c12);
+  p = _mm512_fmadd_ps(p, r, c11);
+  p = _mm512_fmadd_ps(p, r, c10);
+  p = _mm512_fmadd_ps(p, r, c9);
+  p = _mm512_fmadd_ps(p, r, c8);
+  p = _mm512_fmadd_ps(p, r, c7);
+  p = _mm512_fmadd_ps(p, r, c6);
+  p = _mm512_fmadd_ps(p, r, c5);
+  p = _mm512_fmadd_ps(p, r, c4);
+  p = _mm512_fmadd_ps(p, r, c3);
+  p = _mm512_fmadd_ps(p, r, c2);
+  p = _mm512_fmadd_ps(p, r, c1);
+  p = _mm512_fmadd_ps(p, r, c0);
+
+  __m512i ni = _mm512_cvtps_epi32(nf);
+  __m512i biased = _mm512_add_epi32(ni, _mm512_set1_epi32(127));
+  __m512i scaled = _mm512_slli_epi32(biased, 23);
+  __m512 pow2n = _mm512_castsi512_ps(scaled);
+
+  // # 16 MULTs
+  return _mm512_mul_ps(p, pow2n);
+}
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+  (void)pedestrians;
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+  vector<tuple<int, int, float, float, float, float, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+    for (int i = 0; i < n_p; i++) {
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  const int n_p_pad = (n_p + 31) & ~31;
+
+  alignas(64) const float lane_offset_arr[16] = {0.0f, 1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f};
+  const __m512 lane_off = _mm512_load_ps(lane_offset_arr);
+
+  const __m512 v_neg_v0_div_2s = _mm512_set1_ps(NEG_V0_AB_div_TWO_SIGMA);
+  const __m512 v_neg_2s_inv = _mm512_set1_ps(NEG_TWO_SIGMA_INV);
+  const __m512 v_cos_phi_sq = _mm512_set1_ps(cos_phi_sq);
+  const __m512 v_dt_neg = _mm512_set1_ps(dt_neg);
+  const __m512 v_c_neg_dt = _mm512_set1_ps(c_neg_times_dt);
+  const __m512 v_zero = _mm512_setzero_ps();
+  const __m512 v_one = _mm512_set1_ps(1.0f);
+  const __m512 v_eps = _mm512_set1_ps(EPS);
+  const __m512 v_eps_sq = _mm512_set1_ps(EPS_SQ);
+  const __m512 v_np = _mm512_set1_ps((float)n_p);
+  const __m512 v_exp_min = _mm512_set1_ps(-87.0f);
+
+  // # ADDs: n_iter * (
+  // #   n_p * 68 +
+  // #   n_p * n_p_pad * 29
+  // #   n_p * board.n_obstacles * 10
+  // #   n_p_pad * 9
+  // # )
+
+  // # MULTs: n_iter * (
+  // #   n_p * 4 +
+  // #   n_p * n_p_pad * 40
+  // #   n_p * board.n_obstacles * 13
+  // #   n_p_pad * 17
+  // # )
+
+  // # DIVs: n_iter * (
+  // #   n_p * n_p_pad
+  // #   n_p * board.n_obstacles
+  // #   n_p_pad * 2
+  // # )
+
+  // # SQRTs: n_iter * (
+  // #   n_p * n_p_pad * 3
+  // #   n_p * board.n_obstacles
+  // #   n_p_pad * 3
+  // # )
+
+  // # EXPs: n_iter * n_p * board.n_obstacles
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+
+      // # 2 MULTs
+      float desired_velocity_x = e_x[i] * desired_v0[i];
+      float desired_velocity_y = e_y[i] * desired_v0[i];
+
+      // # 2 ADDs, 2 MULTs
+      float f_alpha_dt_x = (desired_velocity_x - curr_vel_x[i]) * dt_times_tau_alpha_inv;
+      float f_alpha_dt_y = (desired_velocity_y - curr_vel_y[i]) * dt_times_tau_alpha_inv;
+
+      const __m512 pix = _mm512_set1_ps(pos_x[i]);
+      const __m512 piy = _mm512_set1_ps(pos_y[i]);
+      const __m512 eix = _mm512_set1_ps(e_x[i]);
+      const __m512 eiy = _mm512_set1_ps(e_y[i]);
+      const __m512 v_i = _mm512_set1_ps((float)i);
+
+      __m512 acc_x_0 = _mm512_setzero_ps();
+      __m512 acc_y_0 = _mm512_setzero_ps();
+      __m512 acc_x_1 = _mm512_setzero_ps();
+      __m512 acc_y_1 = _mm512_setzero_ps();
+
+      for (int j = 0; j < n_p_pad; j += 32) {
+
+        // # 32 ADDs
+        __m512 j_idx_0 = _mm512_add_ps(_mm512_set1_ps((float)j), lane_off);
+        __m512 j_idx_1 = _mm512_add_ps(_mm512_set1_ps((float)j + 16), lane_off);
+
+        __mmask16 m_neq_i_0 = _mm512_cmp_ps_mask(j_idx_0, v_i, _CMP_NEQ_OQ);
+        __mmask16 m_neq_i_1 = _mm512_cmp_ps_mask(j_idx_1, v_i, _CMP_NEQ_OQ);
+
+        __mmask16 m_in_range_0 = _mm512_cmp_ps_mask(j_idx_0, v_np, _CMP_LT_OQ);
+        __mmask16 m_in_range_1 = _mm512_cmp_ps_mask(j_idx_1, v_np, _CMP_LT_OQ);
+
+        __mmask16 active_0 = m_neq_i_0 & m_in_range_0;
+        __mmask16 active_1 = m_neq_i_1 & m_in_range_1;
+
+        __m512 pjx_0 = _mm512_loadu_ps(&pos_x[j]);
+        __m512 pjx_1 = _mm512_loadu_ps(&pos_x[j + 16]);
+
+        __m512 pjy_0 = _mm512_loadu_ps(&pos_y[j]);
+        __m512 pjy_1 = _mm512_loadu_ps(&pos_y[j + 16]);
+
+        __m512 sbe_jx_0 = _mm512_loadu_ps(&sb_e_x[j]);
+        __m512 sbe_jx_1 = _mm512_loadu_ps(&sb_e_x[j + 16]);
+
+        __m512 sbe_jy_0 = _mm512_loadu_ps(&sb_e_y[j]);
+        __m512 sbe_jy_1 = _mm512_loadu_ps(&sb_e_y[j + 16]);
+
+        __m512 sb_jsq_0 = _mm512_loadu_ps(&sb_sq[j]);
+        __m512 sb_jsq_1 = _mm512_loadu_ps(&sb_sq[j + 16]);
+
+        // # 32 ADDs
+        __m512 r_ab_x_0 = _mm512_sub_ps(pix, pjx_0);
+        __m512 r_ab_x_1 = _mm512_sub_ps(pix, pjx_1);
+
+        // # 32 ADDs
+        __m512 r_ab_y_0 = _mm512_sub_ps(piy, pjy_0);
+        __m512 r_ab_y_1 = _mm512_sub_ps(piy, pjy_1);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 r_ab_n2_0 = _mm512_fmadd_ps(r_ab_x_0, r_ab_x_0, _mm512_mul_ps(r_ab_y_0, r_ab_y_0));
+        __m512 r_ab_n2_1 = _mm512_fmadd_ps(r_ab_x_1, r_ab_x_1, _mm512_mul_ps(r_ab_y_1, r_ab_y_1));
+
+        // # 32 SQRTs
+        __m512 r_ab_n_0 = _mm512_sqrt_ps(r_ab_n2_0);
+        __m512 r_ab_n_1 = _mm512_sqrt_ps(r_ab_n2_1);
+
+        // # 32 ADDs
+        __m512 r_ab2_x_0 = _mm512_sub_ps(r_ab_x_0, sbe_jx_0);
+        __m512 r_ab2_x_1 = _mm512_sub_ps(r_ab_x_1, sbe_jx_1);
+
+        // # 32 ADDs
+        __m512 r_ab2_y_0 = _mm512_sub_ps(r_ab_y_0, sbe_jy_0);
+        __m512 r_ab2_y_1 = _mm512_sub_ps(r_ab_y_1, sbe_jy_1);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 r_ab2_n2_0 = _mm512_fmadd_ps(r_ab2_x_0, r_ab2_x_0, _mm512_mul_ps(r_ab2_y_0, r_ab2_y_0));
+        __m512 r_ab2_n2_1 = _mm512_fmadd_ps(r_ab2_x_1, r_ab2_x_1, _mm512_mul_ps(r_ab2_y_1, r_ab2_y_1));
+
+        // # 32 SQRTs
+        __m512 r_ab2_n_0 = _mm512_sqrt_ps(r_ab2_n2_0);
+        __m512 r_ab2_n_1 = _mm512_sqrt_ps(r_ab2_n2_1);
+
+        // # 32 ADDs
+        __m512 r_sum_0 = _mm512_add_ps(r_ab_n_0, r_ab2_n_0);
+        __m512 r_sum_1 = _mm512_add_ps(r_ab_n_1, r_ab2_n_1);
+
+        // # 32 MULTs, 32 ADDs
+        __m512 two_b_sq_0 = _mm512_fmsub_ps(r_sum_0, r_sum_0, sb_jsq_0);
+        __m512 two_b_sq_1 = _mm512_fmsub_ps(r_sum_1, r_sum_1, sb_jsq_1);
+
+        // # 32 SQRTs
+        __m512 two_b_0 = _mm512_sqrt_ps(two_b_sq_0);
+        __m512 two_b_1 = _mm512_sqrt_ps(two_b_sq_1);
+
+        // # 32 MULTs
+        __m512 e_arg_0 = _mm512_mul_ps(two_b_0, v_neg_2s_inv);
+        __m512 e_arg_1 = _mm512_mul_ps(two_b_1, v_neg_2s_inv);
+
+        e_arg_0 = _mm512_max_ps(e_arg_0, v_exp_min);
+        e_arg_1 = _mm512_max_ps(e_arg_1, v_exp_min);
+
+        // # 512 MULTs, 448 ADDs
+        __m512 e_val_0 = _avx512_exp_ps(e_arg_0);
+        __m512 e_val_1 = _avx512_exp_ps(e_arg_1);
+
+        // # 64 MULTs
+        __m512 numer_0 = _mm512_mul_ps(_mm512_mul_ps(r_sum_0, v_neg_v0_div_2s), e_val_0);
+        __m512 numer_1 = _mm512_mul_ps(_mm512_mul_ps(r_sum_1, v_neg_v0_div_2s), e_val_1);
+
+        // # 64 MULTs
+        __m512 denom_0 = _mm512_mul_ps(_mm512_mul_ps(two_b_0, r_ab_n_0), r_ab2_n_0);
+        __m512 denom_1 = _mm512_mul_ps(_mm512_mul_ps(two_b_1, r_ab_n_1), r_ab2_n_1);
+
+        __m512 denom_safe_0 = _mm512_mask_blend_ps(active_0, v_one, denom_0);
+        __m512 denom_safe_1 = _mm512_mask_blend_ps(active_1, v_one, denom_1);
+
+        // # 32 DIVs
+        __m512 smc_0 = _mm512_div_ps(numer_0, denom_safe_0);
+        __m512 smc_1 = _mm512_div_ps(numer_1, denom_safe_1);
+
+        // # 32 MULTs
+        __m512 sm_r_ab_0 = _mm512_mul_ps(smc_0, r_ab2_n_0);
+        __m512 sm_r_ab_1 = _mm512_mul_ps(smc_1, r_ab2_n_1);
+
+        // # 32 MULTs
+        __m512 sm_r_ab2_0 = _mm512_mul_ps(smc_0, r_ab_n_0);
+        __m512 sm_r_ab2_1 = _mm512_mul_ps(smc_1, r_ab_n_1);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 f_x_0 = _mm512_fmadd_ps(r_ab2_x_0, sm_r_ab2_0, _mm512_mul_ps(r_ab_x_0, sm_r_ab_0));
+        __m512 f_x_1 = _mm512_fmadd_ps(r_ab2_x_1, sm_r_ab2_1, _mm512_mul_ps(r_ab_x_1, sm_r_ab_1));
+
+        // # 64 MULTs, 32 ADDs
+        __m512 f_y_0 = _mm512_fmadd_ps(r_ab2_y_0, sm_r_ab2_0, _mm512_mul_ps(r_ab_y_0, sm_r_ab_0));
+        __m512 f_y_1 = _mm512_fmadd_ps(r_ab2_y_1, sm_r_ab2_1, _mm512_mul_ps(r_ab_y_1, sm_r_ab_1));
+
+        // # 64 MULTs, 32 ADDs
+        __m512 f_norm_sq_0 = _mm512_fmadd_ps(f_x_0, f_x_0, _mm512_mul_ps(f_y_0, f_y_0));
+        __m512 f_norm_sq_1 = _mm512_fmadd_ps(f_x_1, f_x_1, _mm512_mul_ps(f_y_1, f_y_1));
+
+        // # 32 MULTs
+        __m512 rhs_0 = _mm512_mul_ps(f_norm_sq_0, v_cos_phi_sq);
+        __m512 rhs_1 = _mm512_mul_ps(f_norm_sq_1, v_cos_phi_sq);
+
+        // # 64 MULTs, 32 ADDs
+        __m512 e_dot_f_0 = _mm512_fmadd_ps(eix, f_x_0, _mm512_mul_ps(eiy, f_y_0));
+        __m512 e_dot_f_1 = _mm512_fmadd_ps(eix, f_x_1, _mm512_mul_ps(eiy, f_y_1));
+
+        // # 32 MULTs
+        __m512 e_dot_f_sq_0 = _mm512_mul_ps(e_dot_f_0, e_dot_f_0);
+        __m512 e_dot_f_sq_1 = _mm512_mul_ps(e_dot_f_1, e_dot_f_1);
+
+        __mmask16 c1_0 = _mm512_cmp_ps_mask(e_dot_f_0, v_zero, _CMP_LT_OQ);
+        __mmask16 c1_1 = _mm512_cmp_ps_mask(e_dot_f_1, v_zero, _CMP_LT_OQ);
+
+        __mmask16 c2_0 = _mm512_cmp_ps_mask(e_dot_f_sq_0, rhs_0, _CMP_GT_OQ);
+        __mmask16 c2_1 = _mm512_cmp_ps_mask(e_dot_f_sq_1, rhs_1, _CMP_GT_OQ);
+
+        __mmask16 cond_0 = c1_0 & c2_0;
+        __mmask16 cond_1 = c1_1 & c2_1;
+
+        __m512 w_val_0 = _mm512_mask_blend_ps(cond_0, v_dt_neg, v_c_neg_dt);
+        __m512 w_val_1 = _mm512_mask_blend_ps(cond_1, v_dt_neg, v_c_neg_dt);
+
+        __m512 f_x_safe_0 = _mm512_maskz_mov_ps(active_0, f_x_0);
+        __m512 f_x_safe_1 = _mm512_maskz_mov_ps(active_1, f_x_1);
+
+        __m512 f_y_safe_0 = _mm512_maskz_mov_ps(active_0, f_y_0);
+        __m512 f_y_safe_1 = _mm512_maskz_mov_ps(active_1, f_y_1);
+
+        // # 32 MULTs, 32 ADDs
+        acc_x_0 = _mm512_fmadd_ps(f_x_safe_0, w_val_0, acc_x_0);
+        acc_x_1 = _mm512_fmadd_ps(f_x_safe_1, w_val_1, acc_x_1);
+
+        // # 32 MULTs, 32 ADDs
+        acc_y_0 = _mm512_fmadd_ps(f_y_safe_0, w_val_0, acc_y_0);
+        acc_y_1 = _mm512_fmadd_ps(f_y_safe_1, w_val_1, acc_y_1);
+      }
+
+      // # 64 ADDs
+      f_alpha_dt_x += _mm512_reduce_add_ps(_mm512_add_ps(acc_x_0, acc_x_1));
+      f_alpha_dt_y += _mm512_reduce_add_ps(_mm512_add_ps(acc_y_0, acc_y_1));
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
+        float v1_x = pos_x[i] - board.obstacles[k].start_x;
+        float v1_y = pos_y[i] - board.obstacles[k].start_y;
+        float v2_x = board.obstacles[k].obstacle_v_x;
+        float v2_y = board.obstacles[k].obstacle_v_y;
+
+        // # 3 MULTs, 1 ADD
+        float projection = (v1_x * v2_x + v1_y * v2_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0f)
+          projection = 0.0f;
+        else if (projection > 1.0f)
+          projection = 1.0f;
+
+        // # 4 ADDs, 2 MULTs
+        float r_aB_x = pos_x[i] - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        float r_aB_y = pos_y[i] - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        float r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+        r_aB_norm = r_aB_norm > EPS ? r_aB_norm : EPS;
+
+        // # 4 MULTs, 1 EXP, 1 DIV
+        float scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+        float f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        float f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        float w_val = dt_neg;
+        float e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -e_y[i] : e_y[i]);
+
+        if (e_dot_f < cos_phi)
+          w_val = c_neg_times_dt;
+
+        // # 2 MULTS
+        float f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        float f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        // # 2 ADDs
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      // # 2 ADDs
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p_pad; i += 16) {
+
+      __m512 w_x_i_vec = _mm512_loadu_ps(&w_x[i]);
+      __m512 w_y_i_vec = _mm512_loadu_ps(&w_y[i]);
+      __m512 v_max_i_vec = _mm512_loadu_ps(&v_max[i]);
+      __m512 pos_x_i_vec = _mm512_loadu_ps(&pos_x[i]);
+      __m512 pos_y_i_vec = _mm512_loadu_ps(&pos_y[i]);
+      __m512 destination_y_i_vec = _mm512_loadu_ps(&destination_y[i]);
+      __m512 destination_x_i_vec = _mm512_loadu_ps(&destination_x[i]);
+      __m512 desired_v0_i_vec = _mm512_loadu_ps(&desired_v0[i]);
+
+      __m512i color_i_vec = _mm512_loadu_si512((__m512i *)&color[i]);
+
+      // # 32 MULTs, 16 ADDs
+      __m512 w_norm_sq = _mm512_fmadd_ps(w_x_i_vec, w_x_i_vec, _mm512_mul_ps(w_y_i_vec, w_y_i_vec));
+
+      // # 16 SQRTs, 16 DIVs
+      __m512 w_norm_inv = _mm512_rsqrt14_ps(w_norm_sq);
+
+      // # 16 MULTs
+      __m512 correction_factor = _mm512_min_ps(_mm512_set1_ps(1.0f), _mm512_mul_ps(v_max_i_vec, w_norm_inv));
+
+      // # 32 MULTs
+      __m512 curr_vel_x_i_vec = _mm512_mul_ps(w_x_i_vec, correction_factor);
+      __m512 curr_vel_y_i_vec = _mm512_mul_ps(w_y_i_vec, correction_factor);
+
+      __m512 dt_vec = _mm512_set1_ps(dt);
+
+      // # 32 MULTs, 32 ADDs
+      pos_x_i_vec = _mm512_fmadd_ps(curr_vel_x_i_vec, dt_vec, pos_x_i_vec);
+      pos_y_i_vec = _mm512_fmadd_ps(curr_vel_y_i_vec, dt_vec, pos_y_i_vec);
+
+      __mmask16 cond_1_left = _mm512_cmpeq_epi32_mask(color_i_vec, _mm512_setzero_si512());
+      __mmask16 cond_1_right = _mm512_cmp_ps_mask(pos_x_i_vec, _mm512_set1_ps(BOARD_LENGTH_MINUS_ONE_POINT_FIVE), _CMP_GT_OQ);
+      __mmask16 cond_1 = cond_1_left & cond_1_right;
+
+      pos_x_i_vec = _mm512_mask_blend_ps(cond_1, pos_x_i_vec, _mm512_set1_ps(1.0f));
+      destination_y_i_vec = _mm512_mask_blend_ps(cond_1, destination_y_i_vec, pos_y_i_vec);
+      w_x_i_vec = _mm512_mask_blend_ps(cond_1, w_x_i_vec, desired_v0_i_vec);
+      w_y_i_vec = _mm512_mask_blend_ps(cond_1, w_y_i_vec, _mm512_setzero_ps());
+      curr_vel_x_i_vec = _mm512_mask_blend_ps(cond_1, curr_vel_x_i_vec, w_x_i_vec);
+      curr_vel_y_i_vec = _mm512_mask_blend_ps(cond_1, curr_vel_y_i_vec, w_y_i_vec);
+
+      __mmask16 cond_2_left = _mm512_cmpeq_epi32_mask(color_i_vec, _mm512_set1_epi32(1));
+      __mmask16 cond_2_right = _mm512_cmp_ps_mask(pos_x_i_vec, _mm512_set1_ps(1.5f), _CMP_LT_OQ);
+      __mmask16 cond_2 = cond_2_left & cond_2_right;
+
+      pos_x_i_vec = _mm512_mask_blend_ps(cond_2, pos_x_i_vec, _mm512_set1_ps(BOARD_LENGTH_MINUS_ONE));
+      destination_y_i_vec = _mm512_mask_blend_ps(cond_2, destination_y_i_vec, pos_y_i_vec);
+      __m512 neg_desired_v0 = _mm512_xor_ps(desired_v0_i_vec, _mm512_set1_ps(-0.0f));
+      w_x_i_vec = _mm512_mask_blend_ps(cond_2, w_x_i_vec, neg_desired_v0);
+      w_y_i_vec = _mm512_mask_blend_ps(cond_2, w_y_i_vec, _mm512_setzero_ps());
+      curr_vel_x_i_vec = _mm512_mask_blend_ps(cond_2, curr_vel_x_i_vec, w_x_i_vec);
+      curr_vel_y_i_vec = _mm512_mask_blend_ps(cond_2, curr_vel_y_i_vec, w_y_i_vec);
+
+      pos_y_i_vec = _mm512_max_ps(pos_y_i_vec, _mm512_set1_ps(WALL_EPS));
+      pos_y_i_vec = _mm512_min_ps(pos_y_i_vec, _mm512_set1_ps(BOARD_WIDTH_MINUS_WALL_EPS));
+
+      // # 32 ADDs
+      __m512 ex_vec = _mm512_sub_ps(destination_x_i_vec, pos_x_i_vec);
+      __m512 ey_vec = _mm512_sub_ps(destination_y_i_vec, pos_y_i_vec);
+
+      // # 32 MULTs, 16 ADDs
+      __m512 e_norm_sq = _mm512_fmadd_ps(ex_vec, ex_vec, _mm512_mul_ps(ey_vec, ey_vec));
+
+      // # 16 DIVs, 16 SQRTs
+      __m512 e_norm_inv = _mm512_rsqrt14_ps(e_norm_sq);
+
+      // # 32 MULTs
+      __m512 e_x_i_vec = _mm512_mul_ps(ex_vec, e_norm_inv);
+      __m512 e_y_i_vec = _mm512_mul_ps(ey_vec, e_norm_inv);
+
+      // # 32 MULTs, 16 ADDs
+      __m512 vel_norm_sq = _mm512_fmadd_ps(curr_vel_x_i_vec, curr_vel_x_i_vec, _mm512_mul_ps(curr_vel_y_i_vec, curr_vel_y_i_vec));
+
+      // # 16 MULTs
+      __m512 sb_sq_i_vec = _mm512_mul_ps(vel_norm_sq, _mm512_set1_ps(delta_t_sq));
+
+      // # 16 SQRTs
+      __m512 vel_norm_dt = _mm512_sqrt_ps(sb_sq_i_vec);
+
+      _mm512_storeu_ps(&curr_vel_x[i], curr_vel_x_i_vec);
+      _mm512_storeu_ps(&curr_vel_y[i], curr_vel_y_i_vec);
+      _mm512_storeu_ps(&w_x[i], w_x_i_vec);
+      _mm512_storeu_ps(&w_y[i], w_y_i_vec);
+      _mm512_storeu_ps(&pos_x[i], pos_x_i_vec);
+      _mm512_storeu_ps(&pos_y[i], pos_y_i_vec);
+      _mm512_storeu_ps(&e_x[i], e_x_i_vec);
+      _mm512_storeu_ps(&e_y[i], e_y_i_vec);
+      _mm512_storeu_ps(&destination_y[i], destination_y_i_vec);
+      _mm512_storeu_ps(&sb_sq[i], sb_sq_i_vec);
+
+      // # 32 MULTs
+      _mm512_storeu_ps(&sb_e_x[i], _mm512_mul_ps(e_x_i_vec, vel_norm_dt));
+      _mm512_storeu_ps(&sb_e_y[i], _mm512_mul_ps(e_y_i_vec, vel_norm_dt));
+    }
+
+#if !BENCHMARK
+    for (int i = 0; i < n_p; i++) {
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+    }
+#endif
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix.cpp b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix.cpp
index d682289..a9bbd5c 100644
--- a/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix.cpp
+++ b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix.cpp
@@ -1,5 +1,5 @@
-// # Changes: Replaced sqrt_ps with rsqrt14_ps in the pedestrian loop, similar to @lbaumberger in variant_float_inlining_SoA_AVX2_polyexp_rsqrt.cpp
-// # Predecessor: variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt.cpp
+// # Changes: Added more correction rounds to _avx512_exp_ps to do 30 flops per scalar
+// # Predecessor: variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt.cpp
 
 #include "../headers/board_reduced_float.hpp"
 #include "../headers/constants_float.hpp"
@@ -13,6 +13,8 @@
 #define SEED 42
 #define BOARD_WIDTH 10.0f
 #define BOARD_LENGTH 50.0f
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0f
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5f
 #define WALL_EPS 1e-3f
 #define BOARD_WIDTH_MINUS_WALL_EPS 9.999f
 
@@ -160,19 +162,20 @@ Pedestrian *init_pedestrians(int n_p) {
 
 Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
 
+// # 256 MULTs, 224 ADDs
 static inline __m512 _avx512_exp_ps(__m512 x) {
-  // define own inlined intrinsic function for approximating the exponent function
-  // e^x = 2^n * e^r --> x = n * ln(2) + r
 
   const __m512 log2e = _mm512_set1_ps(1.4426950408889634f);
+
+  // # 16 MULTs
   __m512 nf = _mm512_roundscale_ps(_mm512_mul_ps(x, log2e), _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
 
   const __m512 ln2 = _mm512_set1_ps(0.6931471805599453f);
-  // r = x - n * ln(2)
+
+  // # 16 MULTs, 16 ADDs
   __m512 r = _mm512_fnmadd_ps(nf, ln2, x);
 
-  // approximation of e^r using taylor polynomial
-  // cNUM --> coefficients
+  const __m512 c13 = _mm512_set1_ps(c13_scalar);
   const __m512 c12 = _mm512_set1_ps(c12_scalar);
   const __m512 c11 = _mm512_set1_ps(c11_scalar);
   const __m512 c10 = _mm512_set1_ps(c10_scalar);
@@ -187,8 +190,10 @@ static inline __m512 _avx512_exp_ps(__m512 x) {
   const __m512 c1 = _mm512_set1_ps(c1_scalar);
   const __m512 c0 = _mm512_set1_ps(c0_scalar);
 
-  // apply coefficients
-  __m512 p = c12;
+  __m512 p = c13;
+
+  // #   13 * (16 MULTs, 16 ADDs)
+  // # = 208 MULTs, 208 ADDs
   p = _mm512_fmadd_ps(p, r, c12);
   p = _mm512_fmadd_ps(p, r, c11);
   p = _mm512_fmadd_ps(p, r, c10);
@@ -203,13 +208,12 @@ static inline __m512 _avx512_exp_ps(__m512 x) {
   p = _mm512_fmadd_ps(p, r, c1);
   p = _mm512_fmadd_ps(p, r, c0);
 
-  // use bit shifting for 2^n and bias because float have 23 bit mantissa
   __m512i ni = _mm512_cvtps_epi32(nf);
   __m512i biased = _mm512_add_epi32(ni, _mm512_set1_epi32(127));
   __m512i scaled = _mm512_slli_epi32(biased, 23);
   __m512 pow2n = _mm512_castsi512_ps(scaled);
 
-  // e^x
+  // # 16 MULTs
   return _mm512_mul_ps(p, pow2n);
 }
 
@@ -248,254 +252,227 @@ int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, c
   const __m512 v_np = _mm512_set1_ps((float)n_p);
   const __m512 v_exp_min = _mm512_set1_ps(-87.0f);
 
+  // # ADDs: n_iter * (
+  // #   n_p * 68 +
+  // #   n_p * n_p_pad * 29 +
+  // #   n_p * board.n_obstacles * 10 +
+  // #   n_p_pad * 7
+  // # )
+
+  // # MULTs: n_iter * (
+  // #   n_p * 4 +
+  // #   n_p * n_p_pad * 42 +
+  // #   n_p * board.n_obstacles * 13 +
+  // #   n_p_pad * 16
+  // # )
+
+  // # DIVs: n_iter * (
+  // #   n_p * n_p_pad * 3 +
+  // #   n_p * board.n_obstacles +
+  // #   n_p_pad * 2
+  // # )
+
+  // # SQRTs: n_iter * (
+  // #   n_p * n_p_pad * 3 +
+  // #   n_p * board.n_obstacles +
+  // #   n_p_pad * 3
+  // # )
+
+  // # EXPs: n_iter * n_p * board.n_obstacles
+
   for (int t = 0; t < n_iter; t++) {
     for (int i = 0; i < n_p; i++) {
-      // set the vector for the current pedestrian
 
-      // get desired velocity
+      // # 2 MULTs
       float desired_velocity_x = e_x[i] * desired_v0[i];
       float desired_velocity_y = e_y[i] * desired_v0[i];
 
-      // velocity change accumulator
+      // # 2 ADDs, 2 MULTs
       float f_alpha_dt_x = (desired_velocity_x - curr_vel_x[i]) * dt_times_tau_alpha_inv;
       float f_alpha_dt_y = (desired_velocity_y - curr_vel_y[i]) * dt_times_tau_alpha_inv;
 
-      // pos
       const __m512 pix = _mm512_set1_ps(pos_x[i]);
       const __m512 piy = _mm512_set1_ps(pos_y[i]);
-      // desired direction
       const __m512 eix = _mm512_set1_ps(e_x[i]);
       const __m512 eiy = _mm512_set1_ps(e_y[i]);
-      // broadcast index used by mask to prevent situation pedestrian[i] == pedestrian[j], see m_neq_i_0 and m_neq_i_1
       const __m512 v_i = _mm512_set1_ps((float)i);
 
-      // accumulators --> 16 pedestrians per accumulator
       __m512 acc_x_0 = _mm512_setzero_ps();
       __m512 acc_y_0 = _mm512_setzero_ps();
       __m512 acc_x_1 = _mm512_setzero_ps();
       __m512 acc_y_1 = _mm512_setzero_ps();
 
-      // force between pedestrians
-      // loop unroll factor 32 with 2 vectors
-      // --> each vector has 16 lanes
       for (int j = 0; j < n_p_pad; j += 32) {
 
-        // [j, j+1, j+2, ..., j+15] and [j+16, j+17, j+18, ..., j+31]
+        // # 32 ADDs
         __m512 j_idx_0 = _mm512_add_ps(_mm512_set1_ps((float)j), lane_off);
         __m512 j_idx_1 = _mm512_add_ps(_mm512_set1_ps((float)j + 16), lane_off);
 
-        // lane is active if j != i
-        // [1,1,1,0,1,1,1,...] 0 --> i == j, 1 --> i != j
         __mmask16 m_neq_i_0 = _mm512_cmp_ps_mask(j_idx_0, v_i, _CMP_NEQ_OQ);
         __mmask16 m_neq_i_1 = _mm512_cmp_ps_mask(j_idx_1, v_i, _CMP_NEQ_OQ);
 
-        // lane is active if j < n_p --> valid range
-        // [1,1,1,0,1,1,1,...] 0 --> j >= n_p, 1 --> j < n_p
         __mmask16 m_in_range_0 = _mm512_cmp_ps_mask(j_idx_0, v_np, _CMP_LT_OQ);
         __mmask16 m_in_range_1 = _mm512_cmp_ps_mask(j_idx_1, v_np, _CMP_LT_OQ);
 
-        // bitwise comparison if not valid --> 0
-        // finally check if lane is valid or not --> prevents i == j, and
-        // j >= n_p because of loop unrolling --> we can therefore omit a remainings for loop
         __mmask16 active_0 = m_neq_i_0 & m_in_range_0;
         __mmask16 active_1 = m_neq_i_1 & m_in_range_1;
 
-        // ---
-
-        // load the positions of the other pedestrians x value
-        // [pos[j], pos[j+1], ..., pos[j+15]] and [pos[16], pos[j+17], ..., pos[j+31]] for x and y
         __m512 pjx_0 = _mm512_loadu_ps(&pos_x[j]);
         __m512 pjx_1 = _mm512_loadu_ps(&pos_x[j + 16]);
         __m512 pjy_0 = _mm512_loadu_ps(&pos_y[j]);
         __m512 pjy_1 = _mm512_loadu_ps(&pos_y[j + 16]);
 
-        // ---
-
-        // step vector of j
-        // [(v_j * delta_t * e_j).x, (v_j+1 * delta_t * e_j+1).x, ...]
         __m512 sbe_jx_0 = _mm512_loadu_ps(&sb_e_x[j]);
         __m512 sbe_jx_1 = _mm512_loadu_ps(&sb_e_x[j + 16]);
         __m512 sbe_jy_0 = _mm512_loadu_ps(&sb_e_y[j]);
         __m512 sbe_jy_1 = _mm512_loadu_ps(&sb_e_y[j + 16]);
 
-        // ---
-
-        // squared step length
-        // [j, j+1, ...]
         __m512 sb_jsq_0 = _mm512_loadu_ps(&sb_sq[j]);
         __m512 sb_jsq_1 = _mm512_loadu_ps(&sb_sq[j + 16]);
 
-        // ---
-
-        // distance vector
-        // [j, j+1, ...]
+        // # 64 ADDs
         __m512 r_ab_x_0 = _mm512_sub_ps(pix, pjx_0);
         __m512 r_ab_x_1 = _mm512_sub_ps(pix, pjx_1);
         __m512 r_ab_y_0 = _mm512_sub_ps(piy, pjy_0);
         __m512 r_ab_y_1 = _mm512_sub_ps(piy, pjy_1);
 
-        // |r_ab|^2 needed for rsqrt to prevent 0 values --> x^2 + y^2
+        // # 64 MULTs, 32 ADDs
         __m512 r_ab_n2_0 = _mm512_fmadd_ps(r_ab_x_0, r_ab_x_0, _mm512_mul_ps(r_ab_y_0, r_ab_y_0));
         __m512 r_ab_n2_1 = _mm512_fmadd_ps(r_ab_x_1, r_ab_x_1, _mm512_mul_ps(r_ab_y_1, r_ab_y_1));
 
-        // prevent cases where distance = 0 (=pedestrian on top of each other)
-        // [dist_i_j, dist_i_j+1, ...]
         r_ab_n2_0 = _mm512_max_ps(r_ab_n2_0, v_eps_sq);
         r_ab_n2_1 = _mm512_max_ps(r_ab_n2_1, v_eps_sq);
 
-        // ---
-
-        // 1 / |r_ab| --> 14 bit rsqrt approx. rsqrt --> 1 / x^2 --> rsqrt(x^2) = 1/x
-        // [1 / dist_i_j, 1 / dist_i_j+1, ...]
+        // # 32 DIVs, 32 SQRTs
         __m512 inv_r_ab_n_0 = _mm512_rsqrt14_ps(r_ab_n2_0);
         __m512 inv_r_ab_n_1 = _mm512_rsqrt14_ps(r_ab_n2_1);
 
-        // r_ab^2 * 1 / r_ab = |r_ab| --> normed
-        // [r_i_j, r_i_j+1, ...]
+        // # 32 MULTs
         __m512 r_ab_n_0 = _mm512_mul_ps(r_ab_n2_0, inv_r_ab_n_0);
         __m512 r_ab_n_1 = _mm512_mul_ps(r_ab_n2_1, inv_r_ab_n_1);
 
-        // r_ab - step[j]
-        // j next pos
+        // # 64 ADDs
         __m512 r_ab2_x_0 = _mm512_sub_ps(r_ab_x_0, sbe_jx_0);
         __m512 r_ab2_x_1 = _mm512_sub_ps(r_ab_x_1, sbe_jx_1);
         __m512 r_ab2_y_0 = _mm512_sub_ps(r_ab_y_0, sbe_jy_0);
         __m512 r_ab2_y_1 = _mm512_sub_ps(r_ab_y_1, sbe_jy_1);
 
-        // |r_ab_2| --> needed for rsqrt trick (r_ab - step)
+        // # 64 MULTs, 32 ADDs
         __m512 r_ab2_n2_0 = _mm512_fmadd_ps(r_ab2_x_0, r_ab2_x_0, _mm512_mul_ps(r_ab2_y_0, r_ab2_y_0));
         __m512 r_ab2_n2_1 = _mm512_fmadd_ps(r_ab2_x_1, r_ab2_x_1, _mm512_mul_ps(r_ab2_y_1, r_ab2_y_1));
 
-        // prevent 0 case --> infinity
         r_ab2_n2_0 = _mm512_max_ps(r_ab2_n2_0, v_eps_sq);
         r_ab2_n2_1 = _mm512_max_ps(r_ab2_n2_1, v_eps_sq);
 
-        // same as before use rsqrt with 14bit approx
-        // [1 / |r_ab_2|, 1 / |r_ab_2|, ...]
+        // # 32 DIVs, 32 SQRTs
         __m512 inv_r_ab2_n_0 = _mm512_rsqrt14_ps(r_ab2_n2_0);
         __m512 inv_r_ab2_n_1 = _mm512_rsqrt14_ps(r_ab2_n2_1);
 
-        // |r_ab_2_second_distance_vector| * 1 / |r_ab_2_first_distance_vector|
+        // # 32 MULTs
         __m512 r_ab2_n_0 = _mm512_mul_ps(r_ab2_n2_0, inv_r_ab2_n_0);
         __m512 r_ab2_n_1 = _mm512_mul_ps(r_ab2_n2_1, inv_r_ab2_n_1);
 
-        // r_sum = |r_ab| + |r_ab - step[j]| --> [w_0, w_1, w_2, ..., w_15] and [w_16, ..., w_31]
+        // # 32 ADDs
         __m512 r_sum_0 = _mm512_add_ps(r_ab_n_0, r_ab2_n_0);
         __m512 r_sum_1 = _mm512_add_ps(r_ab_n_1, r_ab2_n_1);
 
-        // 2_b^2 = r_sum^2 - step[j]^2
+        // # 32 MULTs, 32 ADDs
         __m512 two_b_sq_0 = _mm512_fmsub_ps(r_sum_0, r_sum_0, sb_jsq_0);
         __m512 two_b_sq_1 = _mm512_fmsub_ps(r_sum_1, r_sum_1, sb_jsq_1);
 
-        // prevent 0 case
         two_b_sq_0 = _mm512_max_ps(two_b_sq_0, v_eps_sq);
         two_b_sq_1 = _mm512_max_ps(two_b_sq_1, v_eps_sq);
 
-        // 2_b
-        // 1 / 2_b^(2*0.5) --> 1 / 2_b
+        // # 32 DIVs, 32 SQRTs
         __m512 inv_two_b_0 = _mm512_rsqrt14_ps(two_b_sq_0);
         __m512 inv_two_b_1 = _mm512_rsqrt14_ps(two_b_sq_1);
 
-        // 2_b = 2_b^2 * 1 / 2_b
+        // # 32 MULTs
         __m512 two_b_0 = _mm512_mul_ps(two_b_sq_0, inv_two_b_0);
         __m512 two_b_1 = _mm512_mul_ps(two_b_sq_1, inv_two_b_1);
 
-        // ---
-
-        // two_b * -1 / (2*sigma) --> exp
+        // # 32 MULTs
         __m512 e_arg_0 = _mm512_mul_ps(two_b_0, v_neg_2s_inv);
         __m512 e_arg_1 = _mm512_mul_ps(two_b_1, v_neg_2s_inv);
 
-        // exp(x)
-        // when x is small just assume constant value --> faster exp
         e_arg_0 = _mm512_max_ps(e_arg_0, v_exp_min);
         e_arg_1 = _mm512_max_ps(e_arg_1, v_exp_min);
 
-        // custom fast exponent intrinsic using polynomial approximations and some precomputed constants
+        // # 512 MULTs, 448 ADDs
         __m512 e_val_0 = _avx512_exp_ps(e_arg_0);
         __m512 e_val_1 = _avx512_exp_ps(e_arg_1);
 
-        // [−(V0/σ) * e^(−b/σ) * w/(4b), ...] --> scalar
+        // # 96 MULTs
         __m512 numer_inv_two_b_0 = _mm512_mul_ps(_mm512_mul_ps(r_sum_0, v_neg_v0_div_2s), _mm512_mul_ps(e_val_0, inv_two_b_0));
         __m512 numer_inv_two_b_1 = _mm512_mul_ps(_mm512_mul_ps(r_sum_1, v_neg_v0_div_2s), _mm512_mul_ps(e_val_1, inv_two_b_1));
 
-        // scalar * 1 / |r_ab|
+        // # 32 MULTs
         __m512 sm_r_ab_0 = _mm512_mul_ps(numer_inv_two_b_0, inv_r_ab_n_0);
         __m512 sm_r_ab_1 = _mm512_mul_ps(numer_inv_two_b_1, inv_r_ab_n_1);
 
-        // scalar * 1 / |r_ab_2|
+        // # 32 MULTs
         __m512 sm_r_ab2_0 = _mm512_mul_ps(numer_inv_two_b_0, inv_r_ab2_n_0);
         __m512 sm_r_ab2_1 = _mm512_mul_ps(numer_inv_two_b_1, inv_r_ab2_n_1);
 
-        // summed force
+        // # 128 MULTs, 64 ADDs
         __m512 f_x_0 = _mm512_fmadd_ps(r_ab2_x_0, sm_r_ab2_0, _mm512_mul_ps(r_ab_x_0, sm_r_ab_0));
         __m512 f_x_1 = _mm512_fmadd_ps(r_ab2_x_1, sm_r_ab2_1, _mm512_mul_ps(r_ab_x_1, sm_r_ab_1));
         __m512 f_y_0 = _mm512_fmadd_ps(r_ab2_y_0, sm_r_ab2_0, _mm512_mul_ps(r_ab_y_0, sm_r_ab_0));
         __m512 f_y_1 = _mm512_fmadd_ps(r_ab2_y_1, sm_r_ab2_1, _mm512_mul_ps(r_ab_y_1, sm_r_ab_1));
 
-        // using masking to perform --> field of view formula
-        // if (e_dot_f >= 0.0 || e_dot_f_sq <= rhs)
-        //    return dt_neg;
-        // else
-        //    return c_neg_times_dt;
-        //
+        // # 64 MULTs, 32 ADDs
         __m512 f_norm_sq_0 = _mm512_fmadd_ps(f_x_0, f_x_0, _mm512_mul_ps(f_y_0, f_y_0));
         __m512 f_norm_sq_1 = _mm512_fmadd_ps(f_x_1, f_x_1, _mm512_mul_ps(f_y_1, f_y_1));
 
+        // # 32 MULTs
         __m512 rhs_0 = _mm512_mul_ps(f_norm_sq_0, v_cos_phi_sq);
         __m512 rhs_1 = _mm512_mul_ps(f_norm_sq_1, v_cos_phi_sq);
 
+        // # 64 MULTs, 32 ADDs
         __m512 e_dot_f_0 = _mm512_fmadd_ps(eix, f_x_0, _mm512_mul_ps(eiy, f_y_0));
         __m512 e_dot_f_1 = _mm512_fmadd_ps(eix, f_x_1, _mm512_mul_ps(eiy, f_y_1));
 
+        // # 32 MULTs
         __m512 e_dot_f_sq_0 = _mm512_mul_ps(e_dot_f_0, e_dot_f_0);
         __m512 e_dot_f_sq_1 = _mm512_mul_ps(e_dot_f_1, e_dot_f_1);
 
-        // e_i * f < 0 --> behind
         __mmask16 c1_0 = _mm512_cmp_ps_mask(e_dot_f_0, v_zero, _CMP_LT_OQ);
         __mmask16 c1_1 = _mm512_cmp_ps_mask(e_dot_f_1, v_zero, _CMP_LT_OQ);
 
-        // (e_i * f)^2 > cos(phi) --> out of view
         __mmask16 c2_0 = _mm512_cmp_ps_mask(e_dot_f_sq_0, rhs_0, _CMP_GT_OQ);
         __mmask16 c2_1 = _mm512_cmp_ps_mask(e_dot_f_sq_1, rhs_1, _CMP_GT_OQ);
 
-        // get the case (in field of view or not)
-        // 1 = reduced weight, 0 = regular
         __mmask16 cond_0 = c1_0 & c2_0;
         __mmask16 cond_1 = c1_1 & c2_1;
 
-        // apply the correct weight
         __m512 w_val_0 = _mm512_mask_blend_ps(cond_0, v_dt_neg, v_c_neg_dt);
         __m512 w_val_1 = _mm512_mask_blend_ps(cond_1, v_dt_neg, v_c_neg_dt);
 
-        // take care of i == j case by zeroing it out
-        // see active_0 and active_1 at the beginning of this loop
-        // [f.x or 0, ...]
         __m512 f_x_safe_0 = _mm512_maskz_mov_ps(active_0, f_x_0);
         __m512 f_x_safe_1 = _mm512_maskz_mov_ps(active_1, f_x_1);
         __m512 f_y_safe_0 = _mm512_maskz_mov_ps(active_0, f_y_0);
         __m512 f_y_safe_1 = _mm512_maskz_mov_ps(active_1, f_y_1);
 
-        // accumulate force with weight
-        // force_P_total += weight * force_P_other
+        // # 64 MULTs, 64 ADDs
         acc_x_0 = _mm512_fmadd_ps(f_x_safe_0, w_val_0, acc_x_0);
         acc_x_1 = _mm512_fmadd_ps(f_x_safe_1, w_val_1, acc_x_1);
         acc_y_0 = _mm512_fmadd_ps(f_y_safe_0, w_val_0, acc_y_0);
         acc_y_1 = _mm512_fmadd_ps(f_y_safe_1, w_val_1, acc_y_1);
       }
 
-      // scalar force accumulator
+      // # 64 ADDs
       f_alpha_dt_x += _mm512_reduce_add_ps(_mm512_add_ps(acc_x_0, acc_x_1));
       f_alpha_dt_y += _mm512_reduce_add_ps(_mm512_add_ps(acc_y_0, acc_y_1));
 
-      // force between pedestrian and obstacles
-      // no optimizations, because force between pedestrian is the bottleneck of the computation
       for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
         float v1_x = pos_x[i] - board.obstacles[k].start_x;
         float v1_y = pos_y[i] - board.obstacles[k].start_y;
         float v2_x = board.obstacles[k].obstacle_v_x;
         float v2_y = board.obstacles[k].obstacle_v_y;
 
+        // # 3 MULTs, 1 ADD
         float projection = (v1_x * v2_x + v1_y * v2_y) * board.obstacles[k].length_sq_inv;
 
         if (projection < 0.0f)
@@ -503,12 +480,15 @@ int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, c
         else if (projection > 1.0f)
           projection = 1.0f;
 
+        // # 4 ADDs, 2 MULTs
         float r_aB_x = pos_x[i] - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
         float r_aB_y = pos_y[i] - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
 
+        // # 1 SQRT, 2 MULTs, 1 ADD
         float r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
         r_aB_norm = r_aB_norm > EPS ? r_aB_norm : EPS;
 
+        // # 4 MULTs, 1 EXP, 1 DIV
         float scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
         float f_alpha_B_neg_x = r_aB_x * scalar_mult;
         float f_alpha_B_neg_y = r_aB_y * scalar_mult;
@@ -519,21 +499,21 @@ int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, c
         if (e_dot_f < cos_phi)
           w_val = c_neg_times_dt;
 
+        // # 2 MULTS
         float f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
         float f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
 
+        // # 2 ADDs
         f_alpha_dt_x += f_alpha_B_dt_x;
         f_alpha_dt_y += f_alpha_B_dt_y;
       }
 
+      // # 2 ADDs
       w_x[i] += f_alpha_dt_x;
       w_y[i] += f_alpha_dt_y;
     }
 
-    // update each pedestrian
     for (int i = 0; i < n_p_pad; i += 16) {
-      // load 16 pedestrians per iteration with their data
-      // SoA in vectors
       __m512 w_x_i_vec = _mm512_loadu_ps(&w_x[i]);
       __m512 w_y_i_vec = _mm512_loadu_ps(&w_y[i]);
       __m512 v_max_i_vec = _mm512_loadu_ps(&v_max[i]);
@@ -545,20 +525,27 @@ int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, c
 
       __m512i color_i_vec = _mm512_loadu_si512((__m512i *)&color[i]);
 
+      // # 32 MULTs, 16 ADDs
       __m512 w_norm_sq = _mm512_fmadd_ps(w_x_i_vec, w_x_i_vec, _mm512_mul_ps(w_y_i_vec, w_y_i_vec));
+
+      // # 16 SQRTs, 16 DIVs
       __m512 w_norm_inv = _mm512_rsqrt14_ps(w_norm_sq);
+
+      // # 16 MULTs
       __m512 correction_factor = _mm512_min_ps(_mm512_set1_ps(1.0f), _mm512_mul_ps(v_max_i_vec, w_norm_inv));
 
+      // # 32 MULTs
       __m512 curr_vel_x_i_vec = _mm512_mul_ps(w_x_i_vec, correction_factor);
       __m512 curr_vel_y_i_vec = _mm512_mul_ps(w_y_i_vec, correction_factor);
 
       __m512 dt_vec = _mm512_set1_ps(dt);
+
+      // # 32 MULTs, 32 ADDs
       pos_x_i_vec = _mm512_fmadd_ps(curr_vel_x_i_vec, dt_vec, pos_x_i_vec);
       pos_y_i_vec = _mm512_fmadd_ps(curr_vel_y_i_vec, dt_vec, pos_y_i_vec);
 
-      // differentiate whether pedestrian came from left or right using masking
       __mmask16 cond_1_left = _mm512_cmpeq_epi32_mask(color_i_vec, _mm512_setzero_si512());
-      __mmask16 cond_1_right = _mm512_cmp_ps_mask(pos_x_i_vec, _mm512_set1_ps(BOARD_LENGTH - 1.5f), _CMP_GT_OQ);
+      __mmask16 cond_1_right = _mm512_cmp_ps_mask(pos_x_i_vec, _mm512_set1_ps(BOARD_LENGTH_MINUS_ONE_POINT_FIVE), _CMP_GT_OQ);
       __mmask16 cond_1 = cond_1_left & cond_1_right;
 
       pos_x_i_vec = _mm512_mask_blend_ps(cond_1, pos_x_i_vec, _mm512_set1_ps(1.0f));
@@ -572,7 +559,7 @@ int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, c
       __mmask16 cond_2_right = _mm512_cmp_ps_mask(pos_x_i_vec, _mm512_set1_ps(1.5f), _CMP_LT_OQ);
       __mmask16 cond_2 = cond_2_left & cond_2_right;
 
-      pos_x_i_vec = _mm512_mask_blend_ps(cond_2, pos_x_i_vec, _mm512_set1_ps(BOARD_LENGTH - 1.0f));
+      pos_x_i_vec = _mm512_mask_blend_ps(cond_2, pos_x_i_vec, _mm512_set1_ps(BOARD_LENGTH_MINUS_ONE));
       destination_y_i_vec = _mm512_mask_blend_ps(cond_2, destination_y_i_vec, pos_y_i_vec);
       __m512 neg_desired_v0 = _mm512_xor_ps(desired_v0_i_vec, _mm512_set1_ps(-0.0f));
       w_x_i_vec = _mm512_mask_blend_ps(cond_2, w_x_i_vec, neg_desired_v0);
@@ -583,18 +570,29 @@ int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, c
       pos_y_i_vec = _mm512_max_ps(pos_y_i_vec, _mm512_set1_ps(WALL_EPS));
       pos_y_i_vec = _mm512_min_ps(pos_y_i_vec, _mm512_set1_ps(BOARD_WIDTH_MINUS_WALL_EPS));
 
+      // # 32 ADDs
       __m512 ex_vec = _mm512_sub_ps(destination_x_i_vec, pos_x_i_vec);
       __m512 ey_vec = _mm512_sub_ps(destination_y_i_vec, pos_y_i_vec);
+
+      // # 32 MULTs, 16 ADDs
       __m512 e_norm_sq = _mm512_fmadd_ps(ex_vec, ex_vec, _mm512_mul_ps(ey_vec, ey_vec));
+
+      // # 16 SQRTs, 16 DIVs
       __m512 e_norm_inv = _mm512_rsqrt14_ps(e_norm_sq);
+
+      // # 32 MULTs
       __m512 e_x_i_vec = _mm512_mul_ps(ex_vec, e_norm_inv);
       __m512 e_y_i_vec = _mm512_mul_ps(ey_vec, e_norm_inv);
 
+      // # 32 MULTs, 16 ADDs
       __m512 vel_norm_sq = _mm512_fmadd_ps(curr_vel_x_i_vec, curr_vel_x_i_vec, _mm512_mul_ps(curr_vel_y_i_vec, curr_vel_y_i_vec));
+
+      // # 16 MULTs
       __m512 sb_sq_i_vec = _mm512_mul_ps(vel_norm_sq, _mm512_set1_ps(delta_t_sq));
+
+      // # 16 SQRTs
       __m512 vel_norm_dt = _mm512_sqrt_ps(sb_sq_i_vec);
 
-      // write the data
       _mm512_storeu_ps(&curr_vel_x[i], curr_vel_x_i_vec);
       _mm512_storeu_ps(&curr_vel_y[i], curr_vel_y_i_vec);
       _mm512_storeu_ps(&w_x[i], w_x_i_vec);
@@ -605,6 +603,8 @@ int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, c
       _mm512_storeu_ps(&e_y[i], e_y_i_vec);
       _mm512_storeu_ps(&destination_y[i], destination_y_i_vec);
       _mm512_storeu_ps(&sb_sq[i], sb_sq_i_vec);
+
+      // # 32 MULTs
       _mm512_storeu_ps(&sb_e_x[i], _mm512_mul_ps(e_x_i_vec, vel_norm_dt));
       _mm512_storeu_ps(&sb_e_y[i], _mm512_mul_ps(e_y_i_vec, vel_norm_dt));
     }
diff --git a/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved.cpp b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved.cpp
index 5be10ec..fa51a70 100644
--- a/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved.cpp
+++ b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved.cpp
@@ -176,6 +176,7 @@ static inline __m512 _avx512_exp_ps(__m512 x) {
   // # 16 MULTs, 16 ADDs
   __m512 r = _mm512_fnmadd_ps(nf, ln2, x);
 
+  const __m512 c13 = _mm512_set1_ps(c13_scalar);
   const __m512 c12 = _mm512_set1_ps(c12_scalar);
   const __m512 c11 = _mm512_set1_ps(c11_scalar);
   const __m512 c10 = _mm512_set1_ps(c10_scalar);
@@ -190,7 +191,7 @@ static inline __m512 _avx512_exp_ps(__m512 x) {
   const __m512 c1 = _mm512_set1_ps(c1_scalar);
   const __m512 c0 = _mm512_set1_ps(c0_scalar);
 
-  __m512 p = c12;
+  __m512 p = c13;
 
   // #   13 * (16 MULTs, 16 ADDs)
   // # = 208 MULTs, 208 ADDs
diff --git a/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp.cpp b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp.cpp
index eb21dc4..c1cf0b7 100644
--- a/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp.cpp
+++ b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2_rsqrt_morersqrt_expfix_improved_moreilp.cpp
@@ -174,6 +174,7 @@ static inline __m512 _avx512_exp_ps(__m512 x) {
   // # 16 MULTs, 16 ADDs
   __m512 r = _mm512_fnmadd_ps(nf, ln2, x);
 
+  const __m512 c13 = _mm512_set1_ps(c13_scalar);
   const __m512 c12 = _mm512_set1_ps(c12_scalar);
   const __m512 c11 = _mm512_set1_ps(c11_scalar);
   const __m512 c10 = _mm512_set1_ps(c10_scalar);
@@ -188,7 +189,7 @@ static inline __m512 _avx512_exp_ps(__m512 x) {
   const __m512 c1 = _mm512_set1_ps(c1_scalar);
   const __m512 c0 = _mm512_set1_ps(c0_scalar);
 
-  __m512 p = c12;
+  __m512 p = c13;
 
   // #   13 * (16 MULTs, 16 ADDs)
   // # = 208 MULTs, 208 ADDs
diff --git a/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix.cpp b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix.cpp
new file mode 100644
index 0000000..1548d65
--- /dev/null
+++ b/variants/variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_4_expfix.cpp
@@ -0,0 +1,727 @@
+// # Changes: Added ILP with k = 4 to j-loop
+// # Predecessor: variant_float_inlining_SoA_AVX512_polyexp_moreavx_ilp_2.cpp
+
+#include "../headers/board_reduced_float.hpp"
+#include "../headers/constants_float.hpp"
+#include "../headers/pedestrian_reduced_float.hpp"
+
+#include <immintrin.h>
+
+#include <fstream>
+#include <random>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0f
+#define BOARD_LENGTH 50.0f
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0f
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5f
+#define WALL_EPS 1e-3f
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999f
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+alignas(64) vector<float> pos_x, pos_y;
+alignas(64) vector<float> sb_sq;
+alignas(64) vector<float> sb_e_x, sb_e_y;
+alignas(64) vector<float> e_x, e_y;
+alignas(64) vector<float> curr_vel_x, curr_vel_y;
+alignas(64) vector<float> w_x, w_y;
+alignas(64) vector<float> desired_v0;
+alignas(64) vector<float> v_max;
+alignas(64) vector<float> destination_x, destination_y;
+alignas(64) vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<float> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<float> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<float> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<float> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  const int n_p_pad = (n_p + 63) & ~63;
+
+  pos_x.resize(n_p_pad);
+  pos_y.resize(n_p_pad);
+  sb_sq.resize(n_p_pad);
+  sb_e_x.resize(n_p_pad);
+  sb_e_y.resize(n_p_pad);
+  e_x.resize(n_p_pad);
+  e_y.resize(n_p_pad);
+  curr_vel_x.resize(n_p_pad);
+  curr_vel_y.resize(n_p_pad);
+  w_x.resize(n_p_pad);
+  w_y.resize(n_p_pad);
+  desired_v0.resize(n_p_pad);
+  v_max.resize(n_p_pad);
+  destination_x.resize(n_p_pad);
+  destination_y.resize(n_p_pad);
+  color.resize(n_p_pad);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    float x = x_gen_left(gen);
+    float y = y_gen(gen);
+
+    float dest_x = BOARD_LENGTH + 30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    float x = x_gen_right(gen);
+    float y = y_gen(gen);
+
+    float dest_x = -30.0f;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1f);
+
+    float v_m = 1.3f * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int j = n_p; j < n_p_pad; j++) {
+    pos_x[j] = 1e9f;
+    pos_y[j] = 1e9f;
+    sb_sq[j] = 0.0f;
+    sb_e_x[j] = 0.0f;
+    sb_e_y[j] = 0.0f;
+  }
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+// # 256 MULTs, 224 ADDs
+static inline __m512 _avx512_exp_ps(__m512 x) {
+
+  const __m512 log2e = _mm512_set1_ps(1.4426950408889634f);
+
+  // # 16 MULTs
+  __m512 nf = _mm512_roundscale_ps(_mm512_mul_ps(x, log2e), _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+
+  const __m512 ln2 = _mm512_set1_ps(0.6931471805599453f);
+
+  // # 16 MULTs, 16 ADDs
+  __m512 r = _mm512_fnmadd_ps(nf, ln2, x);
+
+  const __m512 c13 = _mm512_set1_ps(c13_scalar);
+  const __m512 c12 = _mm512_set1_ps(c12_scalar);
+  const __m512 c11 = _mm512_set1_ps(c11_scalar);
+  const __m512 c10 = _mm512_set1_ps(c10_scalar);
+  const __m512 c9 = _mm512_set1_ps(c9_scalar);
+  const __m512 c8 = _mm512_set1_ps(c8_scalar);
+  const __m512 c7 = _mm512_set1_ps(c7_scalar);
+  const __m512 c6 = _mm512_set1_ps(c6_scalar);
+  const __m512 c5 = _mm512_set1_ps(c5_scalar);
+  const __m512 c4 = _mm512_set1_ps(c4_scalar);
+  const __m512 c3 = _mm512_set1_ps(c3_scalar);
+  const __m512 c2 = _mm512_set1_ps(c2_scalar);
+  const __m512 c1 = _mm512_set1_ps(c1_scalar);
+  const __m512 c0 = _mm512_set1_ps(c0_scalar);
+
+  __m512 p = c13;
+
+  // #   13 * (16 MULTs, 16 ADDs)
+  // # = 208 MULTs, 208 ADDs
+  p = _mm512_fmadd_ps(p, r, c12);
+  p = _mm512_fmadd_ps(p, r, c11);
+  p = _mm512_fmadd_ps(p, r, c10);
+  p = _mm512_fmadd_ps(p, r, c9);
+  p = _mm512_fmadd_ps(p, r, c8);
+  p = _mm512_fmadd_ps(p, r, c7);
+  p = _mm512_fmadd_ps(p, r, c6);
+  p = _mm512_fmadd_ps(p, r, c5);
+  p = _mm512_fmadd_ps(p, r, c4);
+  p = _mm512_fmadd_ps(p, r, c3);
+  p = _mm512_fmadd_ps(p, r, c2);
+  p = _mm512_fmadd_ps(p, r, c1);
+  p = _mm512_fmadd_ps(p, r, c0);
+
+  __m512i ni = _mm512_cvtps_epi32(nf);
+  __m512i biased = _mm512_add_epi32(ni, _mm512_set1_epi32(127));
+  __m512i scaled = _mm512_slli_epi32(biased, 23);
+  __m512 pow2n = _mm512_castsi512_ps(scaled);
+
+  // # 16 MULTs
+  return _mm512_mul_ps(p, pow2n);
+}
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+  (void)pedestrians;
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+  vector<tuple<int, int, float, float, float, float, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+    for (int i = 0; i < n_p; i++) {
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  const int n_p_pad = (n_p + 63) & ~63;
+
+  alignas(64) const float lane_offset_arr[16] = {0.0f, 1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f, 9.0f, 10.0f, 11.0f, 12.0f, 13.0f, 14.0f, 15.0f};
+  const __m512 lane_off = _mm512_load_ps(lane_offset_arr);
+
+  const __m512 v_neg_v0_div_2s = _mm512_set1_ps(NEG_V0_AB_div_TWO_SIGMA);
+  const __m512 v_neg_2s_inv = _mm512_set1_ps(NEG_TWO_SIGMA_INV);
+  const __m512 v_cos_phi_sq = _mm512_set1_ps(cos_phi_sq);
+  const __m512 v_dt_neg = _mm512_set1_ps(dt_neg);
+  const __m512 v_c_neg_dt = _mm512_set1_ps(c_neg_times_dt);
+  const __m512 v_zero = _mm512_setzero_ps();
+  const __m512 v_one = _mm512_set1_ps(1.0f);
+  const __m512 v_eps = _mm512_set1_ps(EPS);
+  const __m512 v_eps_sq = _mm512_set1_ps(EPS_SQ);
+  const __m512 v_np = _mm512_set1_ps((float)n_p);
+  const __m512 v_exp_min = _mm512_set1_ps(-87.0f);
+
+  // # ADDs: n_iter * (
+  // #   n_p * 132 +
+  // #   n_p * n_p_pad * 29 +
+  // #   n_p * board.n_obstacles * 10 +
+  // #   n_p_pad * 7
+  // # )
+
+  // # MULTs: n_iter * (
+  // #   n_p * 4 +
+  // #   n_p * n_p_pad * 40 +
+  // #   n_p * board.n_obstacles * 13 +
+  // #   n_p_pad * 15
+  // # )
+
+  // # DIVs: n_iter * (
+  // #   n_p * n_p_pad +
+  // #   n_p * board.n_obstacles +
+  // #   n_p_pad * 2
+  // # )
+
+  // # SQRTs: n_iter * (
+  // #   n_p * n_p_pad * 3 +
+  // #   n_p * board.n_obstacles +
+  // #   n_p_pad * 3
+  // # )
+
+  // # EXPs: n_iter * n_p * board.n_obstacles
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+
+      // # 2 MULTs
+      float desired_velocity_x = e_x[i] * desired_v0[i];
+      float desired_velocity_y = e_y[i] * desired_v0[i];
+
+      // # 2 ADDs, 2 MULTs
+      float f_alpha_dt_x = (desired_velocity_x - curr_vel_x[i]) * dt_times_tau_alpha_inv;
+      float f_alpha_dt_y = (desired_velocity_y - curr_vel_y[i]) * dt_times_tau_alpha_inv;
+
+      const __m512 pix = _mm512_set1_ps(pos_x[i]);
+      const __m512 piy = _mm512_set1_ps(pos_y[i]);
+      const __m512 eix = _mm512_set1_ps(e_x[i]);
+      const __m512 eiy = _mm512_set1_ps(e_y[i]);
+      const __m512 v_i = _mm512_set1_ps((float)i);
+
+      __m512 acc_x_0 = _mm512_setzero_ps();
+      __m512 acc_x_1 = _mm512_setzero_ps();
+      __m512 acc_x_2 = _mm512_setzero_ps();
+      __m512 acc_x_3 = _mm512_setzero_ps();
+
+      __m512 acc_y_0 = _mm512_setzero_ps();
+      __m512 acc_y_1 = _mm512_setzero_ps();
+      __m512 acc_y_2 = _mm512_setzero_ps();
+      __m512 acc_y_3 = _mm512_setzero_ps();
+
+      for (int j = 0; j < n_p_pad; j += 64) {
+
+        // # 64 ADDs
+        __m512 j_idx_0 = _mm512_add_ps(_mm512_set1_ps((float)j), lane_off);
+        __m512 j_idx_1 = _mm512_add_ps(_mm512_set1_ps((float)j + 16), lane_off);
+        __m512 j_idx_2 = _mm512_add_ps(_mm512_set1_ps((float)j + 32), lane_off);
+        __m512 j_idx_3 = _mm512_add_ps(_mm512_set1_ps((float)j + 48), lane_off);
+
+        __mmask16 m_neq_i_0 = _mm512_cmp_ps_mask(j_idx_0, v_i, _CMP_NEQ_OQ);
+        __mmask16 m_neq_i_1 = _mm512_cmp_ps_mask(j_idx_1, v_i, _CMP_NEQ_OQ);
+        __mmask16 m_neq_i_2 = _mm512_cmp_ps_mask(j_idx_2, v_i, _CMP_NEQ_OQ);
+        __mmask16 m_neq_i_3 = _mm512_cmp_ps_mask(j_idx_3, v_i, _CMP_NEQ_OQ);
+
+        __mmask16 m_in_range_0 = _mm512_cmp_ps_mask(j_idx_0, v_np, _CMP_LT_OQ);
+        __mmask16 m_in_range_1 = _mm512_cmp_ps_mask(j_idx_1, v_np, _CMP_LT_OQ);
+        __mmask16 m_in_range_2 = _mm512_cmp_ps_mask(j_idx_2, v_np, _CMP_LT_OQ);
+        __mmask16 m_in_range_3 = _mm512_cmp_ps_mask(j_idx_3, v_np, _CMP_LT_OQ);
+
+        __mmask16 active_0 = m_neq_i_0 & m_in_range_0;
+        __mmask16 active_1 = m_neq_i_1 & m_in_range_1;
+        __mmask16 active_2 = m_neq_i_2 & m_in_range_2;
+        __mmask16 active_3 = m_neq_i_3 & m_in_range_3;
+
+        __m512 pjx_0 = _mm512_loadu_ps(&pos_x[j]);
+        __m512 pjx_1 = _mm512_loadu_ps(&pos_x[j + 16]);
+        __m512 pjx_2 = _mm512_loadu_ps(&pos_x[j + 32]);
+        __m512 pjx_3 = _mm512_loadu_ps(&pos_x[j + 48]);
+
+        __m512 pjy_0 = _mm512_loadu_ps(&pos_y[j]);
+        __m512 pjy_1 = _mm512_loadu_ps(&pos_y[j + 16]);
+        __m512 pjy_2 = _mm512_loadu_ps(&pos_y[j + 32]);
+        __m512 pjy_3 = _mm512_loadu_ps(&pos_y[j + 48]);
+
+        __m512 sbe_jx_0 = _mm512_loadu_ps(&sb_e_x[j]);
+        __m512 sbe_jx_1 = _mm512_loadu_ps(&sb_e_x[j + 16]);
+        __m512 sbe_jx_2 = _mm512_loadu_ps(&sb_e_x[j + 32]);
+        __m512 sbe_jx_3 = _mm512_loadu_ps(&sb_e_x[j + 48]);
+
+        __m512 sbe_jy_0 = _mm512_loadu_ps(&sb_e_y[j]);
+        __m512 sbe_jy_1 = _mm512_loadu_ps(&sb_e_y[j + 16]);
+        __m512 sbe_jy_2 = _mm512_loadu_ps(&sb_e_y[j + 32]);
+        __m512 sbe_jy_3 = _mm512_loadu_ps(&sb_e_y[j + 48]);
+
+        __m512 sb_jsq_0 = _mm512_loadu_ps(&sb_sq[j]);
+        __m512 sb_jsq_1 = _mm512_loadu_ps(&sb_sq[j + 16]);
+        __m512 sb_jsq_2 = _mm512_loadu_ps(&sb_sq[j + 32]);
+        __m512 sb_jsq_3 = _mm512_loadu_ps(&sb_sq[j + 48]);
+
+        // # 64 ADDs
+        __m512 r_ab_x_0 = _mm512_sub_ps(pix, pjx_0);
+        __m512 r_ab_x_1 = _mm512_sub_ps(pix, pjx_1);
+        __m512 r_ab_x_2 = _mm512_sub_ps(pix, pjx_2);
+        __m512 r_ab_x_3 = _mm512_sub_ps(pix, pjx_3);
+
+        // # 64 ADDs
+        __m512 r_ab_y_0 = _mm512_sub_ps(piy, pjy_0);
+        __m512 r_ab_y_1 = _mm512_sub_ps(piy, pjy_1);
+        __m512 r_ab_y_2 = _mm512_sub_ps(piy, pjy_2);
+        __m512 r_ab_y_3 = _mm512_sub_ps(piy, pjy_3);
+
+        // # 128 MULTs, 64 ADDs
+        __m512 r_ab_n2_0 = _mm512_fmadd_ps(r_ab_x_0, r_ab_x_0, _mm512_mul_ps(r_ab_y_0, r_ab_y_0));
+        __m512 r_ab_n2_1 = _mm512_fmadd_ps(r_ab_x_1, r_ab_x_1, _mm512_mul_ps(r_ab_y_1, r_ab_y_1));
+        __m512 r_ab_n2_2 = _mm512_fmadd_ps(r_ab_x_2, r_ab_x_2, _mm512_mul_ps(r_ab_y_2, r_ab_y_2));
+        __m512 r_ab_n2_3 = _mm512_fmadd_ps(r_ab_x_3, r_ab_x_3, _mm512_mul_ps(r_ab_y_3, r_ab_y_3));
+
+        // # 64 SQRTs
+        __m512 r_ab_n_0 = _mm512_sqrt_ps(r_ab_n2_0);
+        __m512 r_ab_n_1 = _mm512_sqrt_ps(r_ab_n2_1);
+        __m512 r_ab_n_2 = _mm512_sqrt_ps(r_ab_n2_2);
+        __m512 r_ab_n_3 = _mm512_sqrt_ps(r_ab_n2_3);
+
+        // # 64 ADDs
+        __m512 r_ab2_x_0 = _mm512_sub_ps(r_ab_x_0, sbe_jx_0);
+        __m512 r_ab2_x_1 = _mm512_sub_ps(r_ab_x_1, sbe_jx_1);
+        __m512 r_ab2_x_2 = _mm512_sub_ps(r_ab_x_2, sbe_jx_2);
+        __m512 r_ab2_x_3 = _mm512_sub_ps(r_ab_x_3, sbe_jx_3);
+
+        // # 64 ADDs
+        __m512 r_ab2_y_0 = _mm512_sub_ps(r_ab_y_0, sbe_jy_0);
+        __m512 r_ab2_y_1 = _mm512_sub_ps(r_ab_y_1, sbe_jy_1);
+        __m512 r_ab2_y_2 = _mm512_sub_ps(r_ab_y_2, sbe_jy_2);
+        __m512 r_ab2_y_3 = _mm512_sub_ps(r_ab_y_3, sbe_jy_3);
+
+        // # 128 MULTs, 64 ADDs
+        __m512 r_ab2_n2_0 = _mm512_fmadd_ps(r_ab2_x_0, r_ab2_x_0, _mm512_mul_ps(r_ab2_y_0, r_ab2_y_0));
+        __m512 r_ab2_n2_1 = _mm512_fmadd_ps(r_ab2_x_1, r_ab2_x_1, _mm512_mul_ps(r_ab2_y_1, r_ab2_y_1));
+        __m512 r_ab2_n2_2 = _mm512_fmadd_ps(r_ab2_x_2, r_ab2_x_2, _mm512_mul_ps(r_ab2_y_2, r_ab2_y_2));
+        __m512 r_ab2_n2_3 = _mm512_fmadd_ps(r_ab2_x_3, r_ab2_x_3, _mm512_mul_ps(r_ab2_y_3, r_ab2_y_3));
+
+        // # 64 SQRTs
+        __m512 r_ab2_n_0 = _mm512_sqrt_ps(r_ab2_n2_0);
+        __m512 r_ab2_n_1 = _mm512_sqrt_ps(r_ab2_n2_1);
+        __m512 r_ab2_n_2 = _mm512_sqrt_ps(r_ab2_n2_2);
+        __m512 r_ab2_n_3 = _mm512_sqrt_ps(r_ab2_n2_3);
+
+        // # 64 ADDs
+        __m512 r_sum_0 = _mm512_add_ps(r_ab_n_0, r_ab2_n_0);
+        __m512 r_sum_1 = _mm512_add_ps(r_ab_n_1, r_ab2_n_1);
+        __m512 r_sum_2 = _mm512_add_ps(r_ab_n_2, r_ab2_n_2);
+        __m512 r_sum_3 = _mm512_add_ps(r_ab_n_3, r_ab2_n_3);
+
+        // # 64 MULTs, 64 ADDs
+        __m512 two_b_sq_0 = _mm512_fmsub_ps(r_sum_0, r_sum_0, sb_jsq_0);
+        __m512 two_b_sq_1 = _mm512_fmsub_ps(r_sum_1, r_sum_1, sb_jsq_1);
+        __m512 two_b_sq_2 = _mm512_fmsub_ps(r_sum_2, r_sum_2, sb_jsq_2);
+        __m512 two_b_sq_3 = _mm512_fmsub_ps(r_sum_3, r_sum_3, sb_jsq_3);
+
+        // # 64 SQRTs
+        __m512 two_b_0 = _mm512_sqrt_ps(two_b_sq_0);
+        __m512 two_b_1 = _mm512_sqrt_ps(two_b_sq_1);
+        __m512 two_b_2 = _mm512_sqrt_ps(two_b_sq_2);
+        __m512 two_b_3 = _mm512_sqrt_ps(two_b_sq_3);
+
+        // # 64 MULTs
+        __m512 e_arg_0 = _mm512_mul_ps(two_b_0, v_neg_2s_inv);
+        __m512 e_arg_1 = _mm512_mul_ps(two_b_1, v_neg_2s_inv);
+        __m512 e_arg_2 = _mm512_mul_ps(two_b_2, v_neg_2s_inv);
+        __m512 e_arg_3 = _mm512_mul_ps(two_b_3, v_neg_2s_inv);
+
+        e_arg_0 = _mm512_max_ps(e_arg_0, v_exp_min);
+        e_arg_1 = _mm512_max_ps(e_arg_1, v_exp_min);
+        e_arg_2 = _mm512_max_ps(e_arg_2, v_exp_min);
+        e_arg_3 = _mm512_max_ps(e_arg_3, v_exp_min);
+
+        // # 1024 MULTs, 896 ADDs
+        __m512 e_val_0 = _avx512_exp_ps(e_arg_0);
+        __m512 e_val_1 = _avx512_exp_ps(e_arg_1);
+        __m512 e_val_2 = _avx512_exp_ps(e_arg_2);
+        __m512 e_val_3 = _avx512_exp_ps(e_arg_3);
+
+        // # 128 MULTs
+        __m512 numer_0 = _mm512_mul_ps(_mm512_mul_ps(r_sum_0, v_neg_v0_div_2s), e_val_0);
+        __m512 numer_1 = _mm512_mul_ps(_mm512_mul_ps(r_sum_1, v_neg_v0_div_2s), e_val_1);
+        __m512 numer_2 = _mm512_mul_ps(_mm512_mul_ps(r_sum_2, v_neg_v0_div_2s), e_val_2);
+        __m512 numer_3 = _mm512_mul_ps(_mm512_mul_ps(r_sum_3, v_neg_v0_div_2s), e_val_3);
+
+        // # 128 MULTs
+        __m512 denom_0 = _mm512_mul_ps(_mm512_mul_ps(two_b_0, r_ab_n_0), r_ab2_n_0);
+        __m512 denom_1 = _mm512_mul_ps(_mm512_mul_ps(two_b_1, r_ab_n_1), r_ab2_n_1);
+        __m512 denom_2 = _mm512_mul_ps(_mm512_mul_ps(two_b_2, r_ab_n_2), r_ab2_n_2);
+        __m512 denom_3 = _mm512_mul_ps(_mm512_mul_ps(two_b_3, r_ab_n_3), r_ab2_n_3);
+
+        __m512 denom_safe_0 = _mm512_mask_blend_ps(active_0, v_one, denom_0);
+        __m512 denom_safe_1 = _mm512_mask_blend_ps(active_1, v_one, denom_1);
+        __m512 denom_safe_2 = _mm512_mask_blend_ps(active_2, v_one, denom_2);
+        __m512 denom_safe_3 = _mm512_mask_blend_ps(active_3, v_one, denom_3);
+
+        // # 64 DIVs
+        __m512 smc_0 = _mm512_div_ps(numer_0, denom_safe_0);
+        __m512 smc_1 = _mm512_div_ps(numer_1, denom_safe_1);
+        __m512 smc_2 = _mm512_div_ps(numer_2, denom_safe_2);
+        __m512 smc_3 = _mm512_div_ps(numer_3, denom_safe_3);
+
+        // # 64 MULTs
+        __m512 sm_r_ab_0 = _mm512_mul_ps(smc_0, r_ab2_n_0);
+        __m512 sm_r_ab_1 = _mm512_mul_ps(smc_1, r_ab2_n_1);
+        __m512 sm_r_ab_2 = _mm512_mul_ps(smc_2, r_ab2_n_2);
+        __m512 sm_r_ab_3 = _mm512_mul_ps(smc_3, r_ab2_n_3);
+
+        // # 64 MULTs
+        __m512 sm_r_ab2_0 = _mm512_mul_ps(smc_0, r_ab_n_0);
+        __m512 sm_r_ab2_1 = _mm512_mul_ps(smc_1, r_ab_n_1);
+        __m512 sm_r_ab2_2 = _mm512_mul_ps(smc_2, r_ab_n_2);
+        __m512 sm_r_ab2_3 = _mm512_mul_ps(smc_3, r_ab_n_3);
+
+        // # 128 MULTs, 64 ADDs
+        __m512 f_x_0 = _mm512_fmadd_ps(r_ab2_x_0, sm_r_ab2_0, _mm512_mul_ps(r_ab_x_0, sm_r_ab_0));
+        __m512 f_x_1 = _mm512_fmadd_ps(r_ab2_x_1, sm_r_ab2_1, _mm512_mul_ps(r_ab_x_1, sm_r_ab_1));
+        __m512 f_x_2 = _mm512_fmadd_ps(r_ab2_x_2, sm_r_ab2_2, _mm512_mul_ps(r_ab_x_2, sm_r_ab_2));
+        __m512 f_x_3 = _mm512_fmadd_ps(r_ab2_x_3, sm_r_ab2_3, _mm512_mul_ps(r_ab_x_3, sm_r_ab_3));
+
+        // # 128 MULTs, 64 ADDs
+        __m512 f_y_0 = _mm512_fmadd_ps(r_ab2_y_0, sm_r_ab2_0, _mm512_mul_ps(r_ab_y_0, sm_r_ab_0));
+        __m512 f_y_1 = _mm512_fmadd_ps(r_ab2_y_1, sm_r_ab2_1, _mm512_mul_ps(r_ab_y_1, sm_r_ab_1));
+        __m512 f_y_2 = _mm512_fmadd_ps(r_ab2_y_2, sm_r_ab2_2, _mm512_mul_ps(r_ab_y_2, sm_r_ab_2));
+        __m512 f_y_3 = _mm512_fmadd_ps(r_ab2_y_3, sm_r_ab2_3, _mm512_mul_ps(r_ab_y_3, sm_r_ab_3));
+
+        // # 128 MULTs, 64 ADDs
+        __m512 f_norm_sq_0 = _mm512_fmadd_ps(f_x_0, f_x_0, _mm512_mul_ps(f_y_0, f_y_0));
+        __m512 f_norm_sq_1 = _mm512_fmadd_ps(f_x_1, f_x_1, _mm512_mul_ps(f_y_1, f_y_1));
+        __m512 f_norm_sq_2 = _mm512_fmadd_ps(f_x_2, f_x_2, _mm512_mul_ps(f_y_2, f_y_2));
+        __m512 f_norm_sq_3 = _mm512_fmadd_ps(f_x_3, f_x_3, _mm512_mul_ps(f_y_3, f_y_3));
+
+        // # 64 MULTs
+        __m512 rhs_0 = _mm512_mul_ps(f_norm_sq_0, v_cos_phi_sq);
+        __m512 rhs_1 = _mm512_mul_ps(f_norm_sq_1, v_cos_phi_sq);
+        __m512 rhs_2 = _mm512_mul_ps(f_norm_sq_2, v_cos_phi_sq);
+        __m512 rhs_3 = _mm512_mul_ps(f_norm_sq_3, v_cos_phi_sq);
+
+        // # 128 MULTs, 64 ADDs
+        __m512 e_dot_f_0 = _mm512_fmadd_ps(eix, f_x_0, _mm512_mul_ps(eiy, f_y_0));
+        __m512 e_dot_f_1 = _mm512_fmadd_ps(eix, f_x_1, _mm512_mul_ps(eiy, f_y_1));
+        __m512 e_dot_f_2 = _mm512_fmadd_ps(eix, f_x_2, _mm512_mul_ps(eiy, f_y_2));
+        __m512 e_dot_f_3 = _mm512_fmadd_ps(eix, f_x_3, _mm512_mul_ps(eiy, f_y_3));
+
+        // # 64 MULTs
+        __m512 e_dot_f_sq_0 = _mm512_mul_ps(e_dot_f_0, e_dot_f_0);
+        __m512 e_dot_f_sq_1 = _mm512_mul_ps(e_dot_f_1, e_dot_f_1);
+        __m512 e_dot_f_sq_2 = _mm512_mul_ps(e_dot_f_2, e_dot_f_2);
+        __m512 e_dot_f_sq_3 = _mm512_mul_ps(e_dot_f_3, e_dot_f_3);
+
+        __mmask16 c1_0 = _mm512_cmp_ps_mask(e_dot_f_0, v_zero, _CMP_LT_OQ);
+        __mmask16 c1_1 = _mm512_cmp_ps_mask(e_dot_f_1, v_zero, _CMP_LT_OQ);
+        __mmask16 c1_2 = _mm512_cmp_ps_mask(e_dot_f_2, v_zero, _CMP_LT_OQ);
+        __mmask16 c1_3 = _mm512_cmp_ps_mask(e_dot_f_3, v_zero, _CMP_LT_OQ);
+
+        __mmask16 c2_0 = _mm512_cmp_ps_mask(e_dot_f_sq_0, rhs_0, _CMP_GT_OQ);
+        __mmask16 c2_1 = _mm512_cmp_ps_mask(e_dot_f_sq_1, rhs_1, _CMP_GT_OQ);
+        __mmask16 c2_2 = _mm512_cmp_ps_mask(e_dot_f_sq_2, rhs_2, _CMP_GT_OQ);
+        __mmask16 c2_3 = _mm512_cmp_ps_mask(e_dot_f_sq_3, rhs_3, _CMP_GT_OQ);
+
+        __mmask16 cond_0 = c1_0 & c2_0;
+        __mmask16 cond_1 = c1_1 & c2_1;
+        __mmask16 cond_2 = c1_2 & c2_2;
+        __mmask16 cond_3 = c1_3 & c2_3;
+
+        __m512 w_val_0 = _mm512_mask_blend_ps(cond_0, v_dt_neg, v_c_neg_dt);
+        __m512 w_val_1 = _mm512_mask_blend_ps(cond_1, v_dt_neg, v_c_neg_dt);
+        __m512 w_val_2 = _mm512_mask_blend_ps(cond_2, v_dt_neg, v_c_neg_dt);
+        __m512 w_val_3 = _mm512_mask_blend_ps(cond_3, v_dt_neg, v_c_neg_dt);
+
+        __m512 f_x_safe_0 = _mm512_maskz_mov_ps(active_0, f_x_0);
+        __m512 f_x_safe_1 = _mm512_maskz_mov_ps(active_1, f_x_1);
+        __m512 f_x_safe_2 = _mm512_maskz_mov_ps(active_2, f_x_2);
+        __m512 f_x_safe_3 = _mm512_maskz_mov_ps(active_3, f_x_3);
+
+        __m512 f_y_safe_0 = _mm512_maskz_mov_ps(active_0, f_y_0);
+        __m512 f_y_safe_1 = _mm512_maskz_mov_ps(active_1, f_y_1);
+        __m512 f_y_safe_2 = _mm512_maskz_mov_ps(active_2, f_y_2);
+        __m512 f_y_safe_3 = _mm512_maskz_mov_ps(active_3, f_y_3);
+
+        // # 64 MULTs, 64 ADDs
+        acc_x_0 = _mm512_fmadd_ps(f_x_safe_0, w_val_0, acc_x_0);
+        acc_x_1 = _mm512_fmadd_ps(f_x_safe_1, w_val_1, acc_x_1);
+        acc_x_2 = _mm512_fmadd_ps(f_x_safe_2, w_val_2, acc_x_2);
+        acc_x_3 = _mm512_fmadd_ps(f_x_safe_3, w_val_3, acc_x_3);
+
+        // # 64 MULTs, 64 ADDs
+        acc_y_0 = _mm512_fmadd_ps(f_y_safe_0, w_val_0, acc_y_0);
+        acc_y_1 = _mm512_fmadd_ps(f_y_safe_1, w_val_1, acc_y_1);
+        acc_y_2 = _mm512_fmadd_ps(f_y_safe_2, w_val_2, acc_y_2);
+        acc_y_3 = _mm512_fmadd_ps(f_y_safe_3, w_val_3, acc_y_3);
+      }
+      // # 128 ADDs
+      f_alpha_dt_x += _mm512_reduce_add_ps(_mm512_add_ps(_mm512_add_ps(acc_x_0, acc_x_1), _mm512_add_ps(acc_x_2, acc_x_3)));
+      f_alpha_dt_y += _mm512_reduce_add_ps(_mm512_add_ps(_mm512_add_ps(acc_y_0, acc_y_1), _mm512_add_ps(acc_y_2, acc_y_3)));
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
+        float v1_x = pos_x[i] - board.obstacles[k].start_x;
+        float v1_y = pos_y[i] - board.obstacles[k].start_y;
+        float v2_x = board.obstacles[k].obstacle_v_x;
+        float v2_y = board.obstacles[k].obstacle_v_y;
+
+        // # 3 MULTs, 1 ADD
+        float projection = (v1_x * v2_x + v1_y * v2_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0f)
+          projection = 0.0f;
+        else if (projection > 1.0f)
+          projection = 1.0f;
+
+        // # 4 ADDs, 2 MULTs
+        float r_aB_x = pos_x[i] - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        float r_aB_y = pos_y[i] - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        float r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+        r_aB_norm = r_aB_norm > EPS ? r_aB_norm : EPS;
+
+        // # 1 EXP, 4 MULTs, 1 DIV
+        float scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+        float f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        float f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        float w_val = dt_neg;
+        float e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -e_y[i] : e_y[i]);
+
+        if (e_dot_f < cos_phi)
+          w_val = c_neg_times_dt;
+
+        // # 2 MULTs
+        float f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        float f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        // # 2 ADDs
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      // # 2 ADDs
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p_pad; i += 16) {
+
+      __m512 w_x_i_vec = _mm512_loadu_ps(&w_x[i]);
+      __m512 w_y_i_vec = _mm512_loadu_ps(&w_y[i]);
+      __m512 v_max_i_vec = _mm512_loadu_ps(&v_max[i]);
+      __m512 pos_x_i_vec = _mm512_loadu_ps(&pos_x[i]);
+      __m512 pos_y_i_vec = _mm512_loadu_ps(&pos_y[i]);
+      __m512 destination_y_i_vec = _mm512_loadu_ps(&destination_y[i]);
+      __m512 destination_x_i_vec = _mm512_loadu_ps(&destination_x[i]);
+      __m512 desired_v0_i_vec = _mm512_loadu_ps(&desired_v0[i]);
+
+      __m512i color_i_vec = _mm512_loadu_si512((__m512i *)&color[i]);
+
+      // # 32 MULTs, 16 ADDs
+      __m512 w_norm_sq = _mm512_fmadd_ps(w_x_i_vec, w_x_i_vec, _mm512_mul_ps(w_y_i_vec, w_y_i_vec));
+
+      // # 16 SQRTs
+      __m512 w_norm = _mm512_sqrt_ps(w_norm_sq);
+
+      // # 16 DIVs
+      __m512 correction_factor = _mm512_min_ps(_mm512_set1_ps(1.0f), _mm512_div_ps(v_max_i_vec, w_norm));
+
+      // # 32 MULTs
+      __m512 curr_vel_x_i_vec = _mm512_mul_ps(w_x_i_vec, correction_factor);
+      __m512 curr_vel_y_i_vec = _mm512_mul_ps(w_y_i_vec, correction_factor);
+
+      __m512 dt_vec = _mm512_set1_ps(dt);
+
+      // # 32 MULTs, 32 ADDs
+      pos_x_i_vec = _mm512_fmadd_ps(curr_vel_x_i_vec, dt_vec, pos_x_i_vec);
+      pos_y_i_vec = _mm512_fmadd_ps(curr_vel_y_i_vec, dt_vec, pos_y_i_vec);
+
+      __mmask16 cond_1_left = _mm512_cmpeq_epi32_mask(color_i_vec, _mm512_setzero_si512());
+      __mmask16 cond_1_right = _mm512_cmp_ps_mask(pos_x_i_vec, _mm512_set1_ps(BOARD_LENGTH_MINUS_ONE_POINT_FIVE), _CMP_GT_OQ);
+      __mmask16 cond_1 = cond_1_left & cond_1_right;
+
+      pos_x_i_vec = _mm512_mask_blend_ps(cond_1, pos_x_i_vec, _mm512_set1_ps(1.0f));
+      destination_y_i_vec = _mm512_mask_blend_ps(cond_1, destination_y_i_vec, pos_y_i_vec);
+      w_x_i_vec = _mm512_mask_blend_ps(cond_1, w_x_i_vec, desired_v0_i_vec);
+      w_y_i_vec = _mm512_mask_blend_ps(cond_1, w_y_i_vec, _mm512_setzero_ps());
+      curr_vel_x_i_vec = _mm512_mask_blend_ps(cond_1, curr_vel_x_i_vec, w_x_i_vec);
+      curr_vel_y_i_vec = _mm512_mask_blend_ps(cond_1, curr_vel_y_i_vec, w_y_i_vec);
+
+      __mmask16 cond_2_left = _mm512_cmpeq_epi32_mask(color_i_vec, _mm512_set1_epi32(1));
+      __mmask16 cond_2_right = _mm512_cmp_ps_mask(pos_x_i_vec, _mm512_set1_ps(1.5f), _CMP_LT_OQ);
+      __mmask16 cond_2 = cond_2_left & cond_2_right;
+
+      pos_x_i_vec = _mm512_mask_blend_ps(cond_2, pos_x_i_vec, _mm512_set1_ps(BOARD_LENGTH_MINUS_ONE));
+      destination_y_i_vec = _mm512_mask_blend_ps(cond_2, destination_y_i_vec, pos_y_i_vec);
+      __m512 neg_desired_v0 = _mm512_xor_ps(desired_v0_i_vec, _mm512_set1_ps(-0.0f));
+      w_x_i_vec = _mm512_mask_blend_ps(cond_2, w_x_i_vec, neg_desired_v0);
+      w_y_i_vec = _mm512_mask_blend_ps(cond_2, w_y_i_vec, _mm512_setzero_ps());
+      curr_vel_x_i_vec = _mm512_mask_blend_ps(cond_2, curr_vel_x_i_vec, w_x_i_vec);
+      curr_vel_y_i_vec = _mm512_mask_blend_ps(cond_2, curr_vel_y_i_vec, w_y_i_vec);
+
+      pos_y_i_vec = _mm512_max_ps(pos_y_i_vec, _mm512_set1_ps(WALL_EPS));
+      pos_y_i_vec = _mm512_min_ps(pos_y_i_vec, _mm512_set1_ps(BOARD_WIDTH_MINUS_WALL_EPS));
+
+      // # 32 ADDs
+      __m512 ex_vec = _mm512_sub_ps(destination_x_i_vec, pos_x_i_vec);
+      __m512 ey_vec = _mm512_sub_ps(destination_y_i_vec, pos_y_i_vec);
+
+      // # 32 MULTs, 16 ADDs
+      __m512 e_norm_sq = _mm512_fmadd_ps(ex_vec, ex_vec, _mm512_mul_ps(ey_vec, ey_vec));
+
+      // # 16 SQRTs
+      __m512 e_norm = _mm512_sqrt_ps(e_norm_sq);
+
+      // # 16 DIVs
+      __m512 e_norm_inv = _mm512_div_ps(_mm512_set1_ps(1.0f), e_norm);
+
+      // # 32 MULTs
+      __m512 e_x_i_vec = _mm512_mul_ps(ex_vec, e_norm_inv);
+      __m512 e_y_i_vec = _mm512_mul_ps(ey_vec, e_norm_inv);
+
+      // # 32 MULTs, 16 ADDs
+      __m512 vel_norm_sq = _mm512_fmadd_ps(curr_vel_x_i_vec, curr_vel_x_i_vec, _mm512_mul_ps(curr_vel_y_i_vec, curr_vel_y_i_vec));
+
+      // # 16 MULTs
+      __m512 sb_sq_i_vec = _mm512_mul_ps(vel_norm_sq, _mm512_set1_ps(delta_t_sq));
+
+      // # 16 SQRTs
+      __m512 vel_norm_dt = _mm512_sqrt_ps(sb_sq_i_vec);
+
+      _mm512_storeu_ps(&curr_vel_x[i], curr_vel_x_i_vec);
+      _mm512_storeu_ps(&curr_vel_y[i], curr_vel_y_i_vec);
+      _mm512_storeu_ps(&w_x[i], w_x_i_vec);
+      _mm512_storeu_ps(&w_y[i], w_y_i_vec);
+      _mm512_storeu_ps(&pos_x[i], pos_x_i_vec);
+      _mm512_storeu_ps(&pos_y[i], pos_y_i_vec);
+      _mm512_storeu_ps(&e_x[i], e_x_i_vec);
+      _mm512_storeu_ps(&e_y[i], e_y_i_vec);
+      _mm512_storeu_ps(&destination_y[i], destination_y_i_vec);
+      _mm512_storeu_ps(&sb_sq[i], sb_sq_i_vec);
+
+      // # 32 MULTs
+      _mm512_storeu_ps(&sb_e_x[i], _mm512_mul_ps(e_x_i_vec, vel_norm_dt));
+      _mm512_storeu_ps(&sb_e_y[i], _mm512_mul_ps(e_y_i_vec, vel_norm_dt));
+    }
+
+#if !BENCHMARK
+    for (int i = 0; i < n_p; i++) {
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+    }
+#endif
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_float_inlining_SoA_flopcount.cpp b/variants/variant_float_inlining_SoA_flopcount.cpp
new file mode 100644
index 0000000..1842150
--- /dev/null
+++ b/variants/variant_float_inlining_SoA_flopcount.cpp
@@ -0,0 +1,380 @@
+// # Changes: SoA layout for pedestrians (file-scope arrays instead of struct fields)
+// # Predecessor: variant_float_inlining.cpp
+
+#include "../headers/board_reduced_float.hpp"
+#include "../headers/constants_float.hpp"
+#include "../headers/pedestrian_reduced_float.hpp"
+
+#include <algorithm>
+#include <cmath>
+#include <fstream>
+#include <iostream>
+#include <math.h>
+#include <random>
+#include <tuple>
+#include <vector>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0f
+#define BOARD_LENGTH 50.0f
+#define WALL_EPS 1e-3f
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999f
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+// we wanna use arrays so we can implement SoA, one array per pedestrian field
+vector<float> pos_x, pos_y;
+vector<float> sb_sq;
+vector<float> sb_e_x, sb_e_y;
+vector<float> e_x, e_y;
+vector<float> curr_vel_x, curr_vel_y;
+vector<float> w_x, w_y;
+vector<float> desired_v0;
+vector<float> v_max;
+vector<float> destination_x, destination_y;
+vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<float> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<float> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<float> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<float> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  // resize all arrays for this run
+  pos_x.resize(n_p);
+  pos_y.resize(n_p);
+  sb_sq.resize(n_p);
+  sb_e_x.resize(n_p);
+  sb_e_y.resize(n_p);
+  e_x.resize(n_p);
+  e_y.resize(n_p);
+  curr_vel_x.resize(n_p);
+  curr_vel_y.resize(n_p);
+  w_x.resize(n_p);
+  w_y.resize(n_p);
+  desired_v0.resize(n_p);
+  v_max.resize(n_p);
+  destination_x.resize(n_p);
+  destination_y.resize(n_p);
+  color.resize(n_p);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    float x = x_gen_left(gen);
+    float y = y_gen(gen);
+
+    float dest_x = BOARD_LENGTH + 30.0; // renamed to avoid shadowing destination_x[]
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1);
+
+    float v_m = 1.3 * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    // Fields the Pedestrian constructor would have zero-initialized (c_v_x=c_v_y=0):
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    float x = x_gen_right(gen);
+    float y = y_gen(gen);
+
+    float dest_x = -30;
+    float dest_y = y;
+
+    float init_e_x = dest_x - x;
+    float init_e_y = dest_y - y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1);
+
+    float v_m = 1.3 * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0f;
+    w_y[i] = 0.0f;
+    curr_vel_x[i] = 0.0f;
+    curr_vel_y[i] = 0.0f;
+    sb_sq[i] = 0.0f;
+    sb_e_x[i] = 0.0f;
+    sb_e_y[i] = 0.0f;
+  }
+
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+  (void)pedestrians;
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+
+  vector<tuple<int, int, float, float, float, float, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+
+    for (int i = 0; i < n_p; i++) {
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  // # ADDs: n_iter * (
+  // #   n_p * 4 +
+  // #   n_p * (n_p - 1) * 14 +
+  // #   n_p * board.n_obstacles * 10 +
+  // #   n_p * 7
+  // # )
+
+  // # MULTs: n_iter * (
+  // #   n_p * 4
+  // #   n_p * (n_p - 1) * 24 +
+  // #   n_p * board.n_obstacles * 13 +
+  // #   n_p * 17
+  // # )
+
+  // # DIVs: n_iter * (
+  // #   n_p * (n_p - 1) +
+  // #   n_p * board.n_obstacles +
+  // #   n_p * 2
+  // # )
+
+  // # SQRTs: n_iter * (
+  // #   n_p * (n_p - 1) * 3 +
+  // #   n_p * board.n_obstacles +
+  // #   n_p * 3
+  // # )
+
+  // # EXPs: n_iter * (
+  // #   n_p * (n_p - 1) +
+  // #   n_p * board.n_obstacles
+  // # )
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+
+      float desired_velocity_x = e_x[i] * desired_v0[i];
+      float desired_velocity_y = e_y[i] * desired_v0[i];
+      float f_alpha_dt_x = (desired_velocity_x - curr_vel_x[i]) * dt_times_tau_alpha_inv;
+      float f_alpha_dt_y = (desired_velocity_y - curr_vel_y[i]) * dt_times_tau_alpha_inv;
+
+      for (int j = 0; j < n_p; j++) {
+        if (i == j)
+          continue;
+
+        float r_ab_x = pos_x[i] - pos_x[j];
+        float r_ab_y = pos_y[i] - pos_y[j];
+
+        float r_ab_norm = sqrt(r_ab_x * r_ab_x + r_ab_y * r_ab_y);
+        r_ab_norm = r_ab_norm > EPS ? r_ab_norm : EPS;
+
+        float r_ab2_x = r_ab_x - sb_e_x[j];
+        float r_ab2_y = r_ab_y - sb_e_y[j];
+
+        float r_ab2_norm = sqrt(r_ab2_x * r_ab2_x + r_ab2_y * r_ab2_y);
+        r_ab2_norm = r_ab2_norm > EPS ? r_ab2_norm : EPS;
+
+        float r_ab_norm_sum = r_ab_norm + r_ab2_norm;
+
+        float two_b_sq = r_ab_norm_sum * r_ab_norm_sum - sb_sq[j];
+        two_b_sq = two_b_sq > EPS_SQ ? two_b_sq : EPS_SQ;
+        float two_b = sqrt(two_b_sq);
+        float scalar_mult_common = (r_ab_norm_sum / (two_b * r_ab_norm * r_ab2_norm)) * (NEG_V0_AB_div_TWO_SIGMA * exp(two_b * NEG_TWO_SIGMA_INV));
+        float scalar_mult_r_ab = scalar_mult_common * r_ab2_norm;
+        float scalar_mult_r_ab2 = scalar_mult_common * r_ab_norm;
+
+        float f_alpha_beta_neg_x = r_ab_x * scalar_mult_r_ab + r_ab2_x * scalar_mult_r_ab2;
+        float f_alpha_beta_neg_y = r_ab_y * scalar_mult_r_ab + r_ab2_y * scalar_mult_r_ab2;
+
+        float w_val;
+
+        float f_norm_sq = f_alpha_beta_neg_x * f_alpha_beta_neg_x + f_alpha_beta_neg_y * f_alpha_beta_neg_y;
+        float rhs = f_norm_sq * cos_phi_sq;
+        float e_dot_f = e_x[i] * f_alpha_beta_neg_x + e_y[i] * f_alpha_beta_neg_y;
+        float e_dot_f_sq = e_dot_f * e_dot_f;
+        if (e_dot_f >= 0.0 || e_dot_f_sq <= rhs)
+          w_val = dt_neg;
+        else
+          w_val = c_neg_times_dt;
+
+        f_alpha_dt_x = f_alpha_dt_x + f_alpha_beta_neg_x * w_val;
+        f_alpha_dt_y = f_alpha_dt_y + f_alpha_beta_neg_y * w_val;
+      }
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        float v1_x = pos_x[i] - board.obstacles[k].start_x;
+        float v1_y = pos_y[i] - board.obstacles[k].start_y;
+        float v2_x = board.obstacles[k].obstacle_v_x;
+        float v2_y = board.obstacles[k].obstacle_v_y;
+
+        float projection = (v1_x * v2_x + v1_y * v2_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0)
+          projection = 0.0;
+        else if (projection > 1.0)
+          projection = 1.0;
+
+        float r_aB_x = pos_x[i] - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        float r_aB_y = pos_y[i] - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        float r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+        r_aB_norm = r_aB_norm > EPS ? r_aB_norm : EPS;
+
+        float scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+        float f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        float f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        float w_val;
+
+        float e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -e_y[i] : e_y[i]);
+
+        if (e_dot_f >= cos_phi)
+          w_val = dt_neg;
+        else
+          w_val = c_neg_times_dt;
+
+        float f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        float f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p; i++) {
+
+      float w_norm = sqrt(w_x[i] * w_x[i] + w_y[i] * w_y[i]);
+
+      float w_norm2 = w_norm > EPS ? w_norm : EPS;
+
+      float correction_factor = min(1.0f, v_max[i] / w_norm2);
+
+      curr_vel_x[i] = w_x[i] * correction_factor;
+      curr_vel_y[i] = w_y[i] * correction_factor;
+
+      pos_x[i] = pos_x[i] + curr_vel_x[i] * dt;
+      pos_y[i] = pos_y[i] + curr_vel_y[i] * dt;
+
+      if (color[i] == 0 && pos_x[i] > (BOARD_LENGTH - 1.5)) {
+        pos_x[i] = 1.0;
+        destination_y[i] = pos_y[i];
+        w_x[i] = desired_v0[i];
+        w_y[i] = 0.0;
+        curr_vel_x[i] = w_x[i];
+        curr_vel_y[i] = w_y[i];
+      }
+      if (color[i] == 1 && pos_x[i] < 1.5) {
+        pos_x[i] = BOARD_LENGTH - 1.0;
+        destination_y[i] = pos_y[i];
+        w_x[i] = -desired_v0[i];
+        w_y[i] = 0.0;
+        curr_vel_x[i] = w_x[i];
+        curr_vel_y[i] = w_y[i];
+      }
+
+      pos_y[i] = clamp(pos_y[i], WALL_EPS, BOARD_WIDTH_MINUS_WALL_EPS);
+
+      float ex = destination_x[i] - pos_x[i];
+      float ey = destination_y[i] - pos_y[i];
+
+      float e_norm = sqrt(ex * ex + ey * ey);
+      float e_norm2 = e_norm > EPS ? e_norm : EPS;
+      float e_norm_inv = 1.0 / e_norm2;
+
+      e_x[i] = ex * e_norm_inv;
+      e_y[i] = ey * e_norm_inv;
+
+      float velocity_norm_sq = curr_vel_x[i] * curr_vel_x[i] + curr_vel_y[i] * curr_vel_y[i];
+
+      float velocity_norm = sqrt(velocity_norm_sq);
+
+      sb_sq[i] = velocity_norm_sq * delta_t_sq;
+
+      sb_e_x[i] = e_x[i] * (velocity_norm * delta_t);
+      sb_e_y[i] = e_y[i] * (velocity_norm * delta_t);
+
+#if !BENCHMARK
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+#endif
+    }
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_float_inlining_flopcount.cpp b/variants/variant_float_inlining_flopcount.cpp
new file mode 100644
index 0000000..482b915
--- /dev/null
+++ b/variants/variant_float_inlining_flopcount.cpp
@@ -0,0 +1,366 @@
+// # Changes: Replaced «double» with «float» in variable declarations
+// # Predecessor: variant_float_baseline.cpp
+
+#include "../headers/board_reduced_float.hpp"
+#include "../headers/constants_float.hpp"
+#include "../headers/pedestrian_reduced_float.hpp"
+
+#include <algorithm>
+#include <cmath>
+#include <fstream>
+#include <iostream>
+#include <math.h>
+#include <random>
+#include <tuple>
+#include <vector>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0f
+#define BOARD_LENGTH 50.0f
+#define WALL_EPS 1e-3f
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999f
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<float> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<float> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<float> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<float> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  Pedestrian *P = new Pedestrian[n_p];
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    float x = x_gen_left(gen);
+    float y = y_gen(gen);
+
+    float start_pos_x = x;
+    float start_pos_y = y;
+
+    float destination_x = BOARD_LENGTH + 30.0;
+    float destination_y = y;
+
+    float init_e_x = destination_x - start_pos_x;
+    float init_e_y = destination_y - start_pos_y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float desired_v0;
+    do {
+      desired_v0 = velocity_gen(gen);
+    } while (desired_v0 <= 0.1);
+
+    float v_max = 1.3 * desired_v0;
+
+    P[i] = Pedestrian(0, 0, 0, start_pos_x, start_pos_y, init_e_unit_x, init_e_unit_y, v_max, desired_v0, destination_x, destination_y);
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    float x = x_gen_right(gen);
+    float y = y_gen(gen);
+
+    float start_pos_x = x;
+    float start_pos_y = y;
+    float destination_x = -30;
+    float destination_y = y;
+
+    float init_e_x = destination_x - start_pos_x;
+    float init_e_y = destination_y - start_pos_y;
+    float init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    float init_e_unit_x = init_e_x / init_e_norm;
+    float init_e_unit_y = init_e_y / init_e_norm;
+
+    float desired_v0;
+    do {
+      desired_v0 = velocity_gen(gen);
+    } while (desired_v0 <= 0.1);
+
+    float v_max = 1.3 * desired_v0;
+
+    P[i] = Pedestrian(1, 0, 0, start_pos_x, start_pos_y, init_e_unit_x, init_e_unit_y, v_max, desired_v0, destination_x, destination_y);
+  }
+
+  return P;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+
+  vector<tuple<int, int, float, float, float, float, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+      samples.emplace_back(0, i, p.curr_pos_x, p.curr_pos_y, p.e_x, p.e_y, p.color);
+    }
+  }
+#endif
+
+  // # ADDs: n_iter * (
+  // #   n_p * 4 +
+  // #   n_p * (n_p - 1) * 14 +
+  // #   n_p * board.n_obstacles * 10 +
+  // #   n_p * 7
+  // # )
+
+  // # MULTs: n_iter * (
+  // #   n_p * 4
+  // #   n_p * (n_p - 1) * 24 +
+  // #   n_p * board.n_obstacles * 13 +
+  // #   n_p * 17
+  // # )
+
+  // # DIVs: n_iter * (
+  // #   n_p * (n_p - 1) +
+  // #   n_p * board.n_obstacles +
+  // #   n_p * 2
+  // # )
+
+  // # SQRTs: n_iter * (
+  // #   n_p * (n_p - 1) * 3 +
+  // #   n_p * board.n_obstacles +
+  // #   n_p * 3
+  // # )
+
+  // # EXPs: n_iter * (
+  // #   n_p * (n_p - 1) +
+  // #   n_p * board.n_obstacles
+  // # )
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+
+      // # 2 MULTs
+      float desired_velocity_x = p.e_x * p.desired_v0;
+      float desired_velocity_y = p.e_y * p.desired_v0;
+
+      // # 2 MULTs, 2 ADDs
+      float f_alpha_dt_x = (desired_velocity_x - p.curr_vel_x) * dt_times_tau_alpha_inv;
+      float f_alpha_dt_y = (desired_velocity_y - p.curr_vel_y) * dt_times_tau_alpha_inv;
+
+      for (int j = 0; j < n_p; j++) {
+        if (i == j)
+          continue;
+
+        // # 2 ADDs
+        float r_ab_x = p.curr_pos_x - pedestrians[j].curr_pos_x;
+        float r_ab_y = p.curr_pos_y - pedestrians[j].curr_pos_y;
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        float r_ab_norm = sqrt(r_ab_x * r_ab_x + r_ab_y * r_ab_y);
+        r_ab_norm = r_ab_norm > EPS ? r_ab_norm : EPS;
+
+        // # 2 ADDs
+        float r_ab2_x = r_ab_x - pedestrians[j].s_b_e_x;
+        float r_ab2_y = r_ab_y - pedestrians[j].s_b_e_y;
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        float r_ab2_norm = sqrt(r_ab2_x * r_ab2_x + r_ab2_y * r_ab2_y);
+        r_ab2_norm = r_ab2_norm > EPS ? r_ab2_norm : EPS;
+
+        // # 1 ADD
+        float r_ab_norm_sum = r_ab_norm + r_ab2_norm;
+
+        // # 1 MULT 1 ADD
+        float two_b_sq = r_ab_norm_sum * r_ab_norm_sum - pedestrians[j].s_b_sq;
+        two_b_sq = two_b_sq > EPS_SQ ? two_b_sq : EPS_SQ;
+
+        // # 1 SQRT
+        float two_b = sqrt(two_b_sq);
+
+        // # 5 MULTs, 1 DIV, 1 EXP
+        float scalar_mult_common = (r_ab_norm_sum / (two_b * r_ab_norm * r_ab2_norm)) * (NEG_V0_AB_div_TWO_SIGMA * exp(two_b * NEG_TWO_SIGMA_INV));
+
+        // # 2 MULTs
+        float scalar_mult_r_ab = scalar_mult_common * r_ab2_norm;
+        float scalar_mult_r_ab2 = scalar_mult_common * r_ab_norm;
+
+        // # 4 MULTs, 2 ADDs
+        float f_alpha_beta_neg_x = r_ab_x * scalar_mult_r_ab + r_ab2_x * scalar_mult_r_ab2;
+        float f_alpha_beta_neg_y = r_ab_y * scalar_mult_r_ab + r_ab2_y * scalar_mult_r_ab2;
+
+        float w_val;
+
+        // # 6 MULTs, 2 ADDs
+        float f_norm_sq = f_alpha_beta_neg_x * f_alpha_beta_neg_x + f_alpha_beta_neg_y * f_alpha_beta_neg_y;
+        float rhs = f_norm_sq * cos_phi_sq;
+        float e_dot_f = p.e_x * f_alpha_beta_neg_x + p.e_y * f_alpha_beta_neg_y;
+        float e_dot_f_sq = e_dot_f * e_dot_f;
+
+        if (e_dot_f >= 0.0 || e_dot_f_sq <= rhs)
+          w_val = dt_neg;
+        else
+          w_val = c_neg_times_dt;
+
+        // # 2 MULTs, 2 ADDs
+        f_alpha_dt_x = f_alpha_dt_x + f_alpha_beta_neg_x * w_val;
+        f_alpha_dt_y = f_alpha_dt_y + f_alpha_beta_neg_y * w_val;
+      }
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
+        float v1_x = p.curr_pos_x - board.obstacles[k].start_x;
+        float v1_y = p.curr_pos_y - board.obstacles[k].start_y;
+        float v2_x = board.obstacles[k].obstacle_v_x;
+        float v2_y = board.obstacles[k].obstacle_v_y;
+
+        // # 3 MULTs, 1 ADD
+        float projection = (v1_x * v2_x + v1_y * v2_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0)
+          projection = 0.0;
+        else if (projection > 1.0)
+          projection = 1.0;
+
+        // # 4 ADDs, 2 MULTs
+        float r_aB_x = p.curr_pos_x - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        float r_aB_y = p.curr_pos_y - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        float r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+        r_aB_norm = r_aB_norm > EPS ? r_aB_norm : EPS;
+
+        // # 4 MULTs, 1 EXP, 1 DIV
+        float scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+        float f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        float f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        float w_val;
+
+        float e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -p.e_y : p.e_y);
+
+        if (e_dot_f >= cos_phi)
+          w_val = dt_neg;
+        else
+          w_val = c_neg_times_dt;
+
+        // # 2 MULTS
+        float f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        float f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        // # 2 ADDs
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      // # 2 ADDs
+      p.w_x += f_alpha_dt_x;
+      p.w_y += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+
+      // # 1 SQRT, 2 MULTs, 1 ADD
+      float w_norm = sqrt(p.w_x * p.w_x + p.w_y * p.w_y);
+
+      float w_norm2 = w_norm > EPS ? w_norm : EPS;
+
+      // # 1 DIV
+      float correction_factor = min(1.0f, p.v_max / w_norm2);
+
+      // # 2 MULTs
+      p.curr_vel_x = p.w_x * correction_factor;
+      p.curr_vel_y = p.w_y * correction_factor;
+
+      // # 2 MULTs, 2 ADDs
+      p.curr_pos_x = p.curr_pos_x + p.curr_vel_x * dt;
+      p.curr_pos_y = p.curr_pos_y + p.curr_vel_y * dt;
+
+      if (p.color == 0 && p.curr_pos_x > (BOARD_LENGTH - 1.5)) {
+        p.curr_pos_x = 1.0;
+        p.destination_y = p.curr_pos_y;
+        p.w_x = p.desired_v0;
+        p.w_y = 0.0;
+        p.curr_vel_x = p.w_x;
+        p.curr_vel_y = p.w_y;
+      }
+      if (p.color == 1 && p.curr_pos_x < 1.5) {
+        p.curr_pos_x = BOARD_LENGTH - 1.0;
+        p.destination_y = p.curr_pos_y;
+        p.w_x = -p.desired_v0;
+        p.w_y = 0.0;
+        p.curr_vel_x = p.w_x;
+        p.curr_vel_y = p.w_y;
+      }
+
+      p.curr_pos_y = clamp(p.curr_pos_y, WALL_EPS, BOARD_WIDTH_MINUS_WALL_EPS);
+
+      // # 2 ADDs
+      float e_x = p.destination_x - p.curr_pos_x;
+      float e_y = p.destination_y - p.curr_pos_y;
+
+      // # 1 SQRT, 2 MULTs, 1 ADD
+      float e_norm = sqrt(e_x * e_x + e_y * e_y);
+
+      float e_norm2 = e_norm > EPS ? e_norm : EPS;
+
+      // # 1 DIV
+      float e_norm_inv = 1.0 / e_norm2;
+
+      // # 2 MULTs
+      p.e_x = e_x * e_norm_inv;
+      p.e_y = e_y * e_norm_inv;
+
+      // # 2 MULTs, 1 ADD
+      float velocity_norm_sq = p.curr_vel_x * p.curr_vel_x + p.curr_vel_y * p.curr_vel_y;
+
+      // # 1 SQRT
+      float velocity_norm = sqrt(velocity_norm_sq);
+
+      // # 5 MULTs
+      p.s_b_sq = velocity_norm_sq * delta_t_sq;
+      p.s_b_e_x = p.e_x * (velocity_norm * delta_t);
+      p.s_b_e_y = p.e_y * (velocity_norm * delta_t);
+
+#if !BENCHMARK
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, p.curr_pos_x, p.curr_pos_y, p.e_x, p.e_y, p.color);
+      }
+#endif
+    }
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_inlining_7_SoA_AVX2_polyexp_expfix.cpp b/variants/variant_inlining_7_SoA_AVX2_polyexp_expfix.cpp
index 36fef1a..a5c93b8 100644
--- a/variants/variant_inlining_7_SoA_AVX2_polyexp_expfix.cpp
+++ b/variants/variant_inlining_7_SoA_AVX2_polyexp_expfix.cpp
@@ -190,21 +190,23 @@ static inline __m256d exp256_pd(__m256d x) {
   const __m256d ln2 = _mm256_set1_pd(0.6931471805599453);
   __m256d r = _mm256_fnmadd_pd(nf, ln2, x); // # 4 MULTs, 4 ADDs
 
-  const __m256d c12 = _mm256_set1_pd(1.0 / 479001600.0);
-  const __m256d c11 = _mm256_set1_pd(1.0 / 39916800.0);
-  const __m256d c10 = _mm256_set1_pd(1.0 / 3628800.0);
-  const __m256d c9  = _mm256_set1_pd(1.0 / 362880.0);
-  const __m256d c8  = _mm256_set1_pd(1.0 / 40320.0);
-  const __m256d c7  = _mm256_set1_pd(1.0 / 5040.0);
-  const __m256d c6  = _mm256_set1_pd(1.0 / 720.0);
-  const __m256d c5  = _mm256_set1_pd(1.0 / 120.0);
-  const __m256d c4  = _mm256_set1_pd(1.0 / 24.0);
-  const __m256d c3  = _mm256_set1_pd(1.0 / 6.0);
-  const __m256d c2  = _mm256_set1_pd(0.5);
-  const __m256d c1  = _mm256_set1_pd(1.0);
-  const __m256d c0  = _mm256_set1_pd(1.0);
+  const __m256d c13 = _mm256_set1_pd(c13_scalar);
+  const __m256d c12 = _mm256_set1_pd(c12_scalar);
+  const __m256d c11 = _mm256_set1_pd(c11_scalar);
+  const __m256d c10 = _mm256_set1_pd(c10_scalar);
+  const __m256d c9 = _mm256_set1_pd(c9_scalar);
+  const __m256d c8 = _mm256_set1_pd(c8_scalar);
+  const __m256d c7 = _mm256_set1_pd(c7_scalar);
+  const __m256d c6 = _mm256_set1_pd(c6_scalar);
+  const __m256d c5 = _mm256_set1_pd(c5_scalar);
+  const __m256d c4 = _mm256_set1_pd(c4_scalar);
+  const __m256d c3 = _mm256_set1_pd(c3_scalar);
+  const __m256d c2 = _mm256_set1_pd(c2_scalar);
+  const __m256d c1 = _mm256_set1_pd(c1_scalar);
+  const __m256d c0 = _mm256_set1_pd(c0_scalar);
+
+  __m256d p = c13;
 
-  __m256d p = c12;
   // #   13 * (4 MULTs, 4 ADDs)
   // # = 52 MULTs, 52 ADDs
   p = _mm256_fmadd_pd(p, r, c12);
diff --git a/variants/variant_inlining_7_SoA_flopcount.cpp b/variants/variant_inlining_7_SoA_flopcount.cpp
new file mode 100644
index 0000000..8c10a6a
--- /dev/null
+++ b/variants/variant_inlining_7_SoA_flopcount.cpp
@@ -0,0 +1,362 @@
+// # This file uses avx2
+// # This is the first variant and should serve as a base for further avx2 optimizations
+
+// Okay our first step is to use a SoA instead of an AoS
+// We only do SoA for the pedestrian struct, there we wanna optimize such that we access for the ped-ped forces adjacent pedestrians (in the arrays)
+
+#include "../headers/board.hpp"
+#include "../headers/constants.hpp"
+#include "../headers/pedestrian_reduced.hpp"
+
+#include <algorithm>
+#include <cmath>
+#include <fstream>
+#include <iostream>
+#include <math.h>
+#include <random>
+#include <tuple>
+#include <vector>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0
+#define BOARD_LENGTH 50.0
+#define WALL_EPS 1e-3
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+// we wanna use arrays so we can implement SOA, we create arrays for all the data we need and access
+vector<double> pos_x, pos_y;
+vector<double> sb_sq;
+vector<double> sb_e_x, sb_e_y;
+vector<double> e_x, e_y;
+vector<double> curr_vel_x, curr_vel_y;
+vector<double> w_x, w_y;
+vector<double> desired_v0;
+vector<double> v_max;
+vector<double> destination_x, destination_y;
+vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<double> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<double> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<double> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<double> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  // resize all our arrays for this run
+  pos_x.resize(n_p);
+  pos_y.resize(n_p);
+  sb_sq.resize(n_p);
+  sb_e_x.resize(n_p);
+  sb_e_y.resize(n_p);
+  e_x.resize(n_p);
+  e_y.resize(n_p);
+  curr_vel_x.resize(n_p);
+  curr_vel_y.resize(n_p);
+  w_x.resize(n_p);
+  w_y.resize(n_p);
+  desired_v0.resize(n_p);
+  v_max.resize(n_p);
+  destination_x.resize(n_p);
+  destination_y.resize(n_p);
+  color.resize(n_p);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    double x = x_gen_left(gen);
+    double y = y_gen(gen);
+
+    double dest_x = BOARD_LENGTH + 30.0; // renamed to avoid shadowing destination_x[]
+    double dest_y = y;                   // same idea for destination_y[]
+
+    double init_e_x = dest_x - x;
+    double init_e_y = dest_y - y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1);
+
+    double v_m = 1.3 * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    // Fields the Pedestrian constructor would have zero-initialized:
+    w_x[i] = 0.0;
+    w_y[i] = 0.0;
+    curr_vel_x[i] = 0.0;
+    curr_vel_y[i] = 0.0;
+    sb_sq[i] = 0.0;
+    sb_e_x[i] = 0.0;
+    sb_e_y[i] = 0.0;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    double x = x_gen_right(gen);
+    double y = y_gen(gen);
+
+    double dest_x = -30;
+    double dest_y = y;
+
+    double init_e_x = dest_x - x;
+    double init_e_y = dest_y - y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1);
+
+    double v_m = 1.3 * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    // Fields the Pedestrian constructor would have zero-initialized:
+    w_x[i] = 0.0;
+    w_y[i] = 0.0;
+    curr_vel_x[i] = 0.0;
+    curr_vel_y[i] = 0.0;
+    sb_sq[i] = 0.0;
+    sb_e_x[i] = 0.0;
+    sb_e_y[i] = 0.0;
+  }
+
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+
+  vector<tuple<int, int, double, double, double, double, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+
+    for (int i = 0; i < n_p; i++) {
+      // Pedestrian &p = pedestrians[i];
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  // # ADDs: n_iter * n_p * 11 +
+  // #        n_iter * n_p * (n_p - 1) * 14 +
+  // #        n_iter * n_p * board.n_obstacles * 12
+
+  // # MULTs: n_iter * n_p * 21 +
+  // #        n_iter * n_p * (n_p - 1) * 24 +
+  // #        n_iter * n_p * board.n_obstacles * 19
+
+  // # DIVs: n_iter * n_p * 2 +
+  // #        n_iter * n_p * (n_p - 1) +
+  // #        n_iter * n_p * board.n_obstacles
+
+  // # SQRTs: n_iter * n_p * 3 +
+  // #        n_iter * n_p * (n_p - 1) * 3 +
+  // #        n_iter * n_p * board.n_obstacles
+
+  // # EXPs: n_iter * n_p * (n_p - 1) +
+  // #       n_iter * n_p * board.n_obstacles
+
+  for (int t = 0; t < n_iter; t++) {
+
+    for (int i = 0; i < n_p; i++) {
+
+      double desired_velocity_x = e_x[i] * desired_v0[i];
+      double desired_velocity_y = e_y[i] * desired_v0[i];
+
+      double f_alpha_dt_x = (desired_velocity_x - curr_vel_x[i]) * dt_times_tau_alpha_inv;
+      double f_alpha_dt_y = (desired_velocity_y - curr_vel_y[i]) * dt_times_tau_alpha_inv;
+
+      for (int j = 0; j < n_p; j++) {
+
+        if (i == j)
+          continue;
+
+        double r_ab_x = pos_x[i] - pos_x[j];
+        double r_ab_y = pos_y[i] - pos_y[j];
+
+        double r_ab_norm = sqrt(r_ab_x * r_ab_x + r_ab_y * r_ab_y);
+
+        double r_ab2_x = r_ab_x - sb_e_x[j];
+        double r_ab2_y = r_ab_y - sb_e_y[j];
+
+        double r_ab2_norm = sqrt(r_ab2_x * r_ab2_x + r_ab2_y * r_ab2_y);
+        double r_ab_norm_sum = r_ab_norm + r_ab2_norm;
+        double two_b_sq = r_ab_norm_sum * r_ab_norm_sum - sb_sq[j];
+        double two_b = sqrt(two_b_sq);
+        double scalar_mult_common = (r_ab_norm_sum / (two_b * r_ab_norm * r_ab2_norm)) * (NEG_V0_AB_div_TWO_SIGMA * exp(two_b * NEG_TWO_SIGMA_INV));
+        double scalar_mult_r_ab = scalar_mult_common * r_ab2_norm;
+        double scalar_mult_r_ab2 = scalar_mult_common * r_ab_norm;
+
+        double f_alpha_beta_neg_x = r_ab_x * scalar_mult_r_ab + r_ab2_x * scalar_mult_r_ab2;
+        double f_alpha_beta_neg_y = r_ab_y * scalar_mult_r_ab + r_ab2_y * scalar_mult_r_ab2;
+
+        double w_val;
+        double a = (e_x[i] * f_alpha_beta_neg_x + e_y[i] * f_alpha_beta_neg_y) * cos_phi_inv;
+        double b_sq = f_alpha_beta_neg_x * f_alpha_beta_neg_x + f_alpha_beta_neg_y * f_alpha_beta_neg_y;
+
+        w_val = dt_neg;
+        if (a >= 0 && (a * a > b_sq)) {
+          w_val = c_neg_times_dt;
+        }
+
+        f_alpha_dt_x = f_alpha_dt_x + f_alpha_beta_neg_x * w_val;
+        f_alpha_dt_y = f_alpha_dt_y + f_alpha_beta_neg_y * w_val;
+      }
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        double v1_x = pos_x[i] - board.obstacles[k].start.x;
+        double v1_y = pos_y[i] - board.obstacles[k].start.y;
+
+        double projection = (v1_x * board.obstacles[k].obstacle_v.x + v1_y * board.obstacles[k].obstacle_v.y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0)
+          projection = 0.0;
+        else if (projection > 1.0)
+          projection = 1.0;
+
+        double r_aB_x = pos_x[i] - (board.obstacles[k].start.x + board.obstacles[k].obstacle_v.x * projection);
+        double r_aB_y = pos_y[i] - (board.obstacles[k].start.y + board.obstacles[k].obstacle_v.y * projection);
+
+        double r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+        double scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+        double f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        double f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        double w_val;
+        double a = (e_x[i] * f_alpha_B_neg_x + e_y[i] * f_alpha_B_neg_y) * cos_phi_inv;
+        double b_sq = f_alpha_B_neg_x * f_alpha_B_neg_x + f_alpha_B_neg_y * f_alpha_B_neg_y;
+
+        w_val = dt_neg;
+        if (a >= 0 && (a * a > b_sq)) {
+          w_val = c_neg_times_dt;
+        }
+
+        double f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        double f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p; i++) {
+
+      double w_norm = sqrt(w_x[i] * w_x[i] + w_y[i] * w_y[i]);
+
+      double correction_factor = min(1.0, v_max[i] / w_norm);
+
+      curr_vel_x[i] = w_x[i] * correction_factor;
+      curr_vel_y[i] = w_y[i] * correction_factor;
+
+      pos_x[i] = pos_x[i] + curr_vel_x[i] * dt;
+      pos_y[i] = pos_y[i] + curr_vel_y[i] * dt;
+
+      if (color[i] == 0 && pos_x[i] > (BOARD_LENGTH - 1.5)) {
+        pos_x[i] = 1.0;
+        destination_y[i] = pos_y[i];
+        w_x[i] = desired_v0[i];
+        w_y[i] = 0.0;
+        curr_vel_x[i] = w_x[i];
+        curr_vel_y[i] = w_y[i];
+      }
+      if (color[i] == 1 && pos_x[i] < 1.5) {
+        pos_x[i] = BOARD_LENGTH - 1.0;
+        destination_y[i] = pos_y[i];
+        w_x[i] = -desired_v0[i];
+        w_y[i] = 0.0;
+        curr_vel_x[i] = w_x[i];
+        curr_vel_y[i] = w_y[i];
+      }
+
+      pos_y[i] = clamp(pos_y[i], WALL_EPS, BOARD_WIDTH_MINUS_WALL_EPS);
+
+      double ex = destination_x[i] - pos_x[i];
+      double ey = destination_y[i] - pos_y[i];
+
+      double e_norm_inv = 1.0 / sqrt(ex * ex + ey * ey);
+
+      e_x[i] = ex * e_norm_inv;
+      e_y[i] = ey * e_norm_inv;
+
+      double velocity_norm_sq = curr_vel_x[i] * curr_vel_x[i] + curr_vel_y[i] * curr_vel_y[i];
+
+      double velocity_norm = sqrt(velocity_norm_sq);
+
+      sb_sq[i] = velocity_norm_sq * delta_t_sq;
+
+      sb_e_x[i] = e_x[i] * (velocity_norm * delta_t);
+      sb_e_y[i] = e_y[i] * (velocity_norm * delta_t);
+
+#if !BENCHMARK
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+#endif
+    }
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_inlining_7_flopcount.cpp b/variants/variant_inlining_7_flopcount.cpp
new file mode 100644
index 0000000..9e5b28a
--- /dev/null
+++ b/variants/variant_inlining_7_flopcount.cpp
@@ -0,0 +1,352 @@
+// # This file performs in-lining for run_simulation.
+// # Changes: Removed 2 sqrt() calls from the loop bodies
+// # Predecessor: variant_inlining_6.cpp
+
+#include "../headers/board.hpp"
+#include "../headers/constants.hpp"
+#include "../headers/pedestrian_reduced.hpp"
+
+#include <algorithm>
+#include <cmath>
+#include <fstream>
+#include <iostream>
+#include <math.h>
+#include <random>
+#include <tuple>
+#include <vector>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0
+#define BOARD_LENGTH 50.0
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5
+#define WALL_EPS 1e-3
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<double> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<double> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<double> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<double> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  Pedestrian *P = new Pedestrian[n_p];
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    double x = x_gen_left(gen);
+    double y = y_gen(gen);
+
+    double start_pos_x = x;
+    double start_pos_y = y;
+
+    double destination_x = BOARD_LENGTH + 30.0;
+    double destination_y = y;
+
+    double init_e_x = destination_x - start_pos_x;
+    double init_e_y = destination_y - start_pos_y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double desired_v0;
+    do {
+      desired_v0 = velocity_gen(gen);
+    } while (desired_v0 <= 0.1);
+
+    double v_max = 1.3 * desired_v0;
+
+    P[i] = Pedestrian(0, 0, 0, start_pos_x, start_pos_y, init_e_unit_x, init_e_unit_y, v_max, desired_v0, destination_x, destination_y);
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    double x = x_gen_right(gen);
+    double y = y_gen(gen);
+
+    double start_pos_x = x;
+    double start_pos_y = y;
+    double destination_x = -30;
+    double destination_y = y;
+
+    double init_e_x = destination_x - start_pos_x;
+    double init_e_y = destination_y - start_pos_y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double desired_v0;
+    do {
+      desired_v0 = velocity_gen(gen);
+    } while (desired_v0 <= 0.1);
+
+    double v_max = 1.3 * desired_v0;
+
+    P[i] = Pedestrian(1, 0, 0, start_pos_x, start_pos_y, init_e_unit_x, init_e_unit_y, v_max, desired_v0, destination_x, destination_y);
+  }
+
+  return P;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+
+  vector<tuple<int, int, double, double, double, double, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+      samples.emplace_back(0, i, p.curr_pos_x, p.curr_pos_y, p.e_x, p.e_y, p.color);
+    }
+  }
+#endif
+
+  // # ADDs: n_iter * n_p * 11 +
+  // #        n_iter * n_p * (n_p - 1) * 14 +
+  // #        n_iter * n_p * board.n_obstacles * 12
+
+  // # MULTs: n_iter * n_p * 21 +
+  // #        n_iter * n_p * (n_p - 1) * 24 +
+  // #        n_iter * n_p * board.n_obstacles * 19
+
+  // # DIVs: n_iter * n_p * 2 +
+  // #        n_iter * n_p * (n_p - 1) +
+  // #        n_iter * n_p * board.n_obstacles
+
+  // # SQRTs: n_iter * n_p * 3 +
+  // #        n_iter * n_p * (n_p - 1) * 3 +
+  // #        n_iter * n_p * board.n_obstacles
+
+  // # EXPs: n_iter * n_p * (n_p - 1) +
+  // #       n_iter * n_p * board.n_obstacles
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+
+      // # 2 MULTs
+      double desired_velocity_x = p.e_x * p.desired_v0;
+      double desired_velocity_y = p.e_y * p.desired_v0;
+
+      // # 2 MULTs, 2 ADDs
+      double f_alpha_dt_x = (desired_velocity_x - p.curr_vel_x) * dt_times_tau_alpha_inv;
+      double f_alpha_dt_y = (desired_velocity_y - p.curr_vel_y) * dt_times_tau_alpha_inv;
+
+      for (int j = 0; j < n_p; j++) {
+        if (i == j)
+          continue;
+
+        // # 2 ADDs
+        double r_ab_x = p.curr_pos_x - pedestrians[j].curr_pos_x;
+        double r_ab_y = p.curr_pos_y - pedestrians[j].curr_pos_y;
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        double r_ab_norm = sqrt(r_ab_x * r_ab_x + r_ab_y * r_ab_y);
+
+        // # 2 ADDs
+        double r_ab2_x = r_ab_x - pedestrians[j].s_b_e_x;
+        double r_ab2_y = r_ab_y - pedestrians[j].s_b_e_y;
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        double r_ab2_norm = sqrt(r_ab2_x * r_ab2_x + r_ab2_y * r_ab2_y);
+
+        // # 1ADD
+        double r_ab_norm_sum = r_ab_norm + r_ab2_norm;
+
+        // # 1 MULT, 1 ADD
+        double two_b_sq = r_ab_norm_sum * r_ab_norm_sum - pedestrians[j].s_b_sq;
+
+        // # 1 SQRT
+        double two_b = sqrt(two_b_sq);
+
+        // # 5 MULTs, 1 EXP, 1 DIV
+        double scalar_mult_common = (r_ab_norm_sum / (two_b * r_ab_norm * r_ab2_norm)) * (NEG_V0_AB_div_TWO_SIGMA * exp(two_b * NEG_TWO_SIGMA_INV));
+
+        // # 2 MULTs
+        double scalar_mult_r_ab = scalar_mult_common * r_ab2_norm;
+        double scalar_mult_r_ab2 = scalar_mult_common * r_ab_norm;
+
+        // # 4 MULTs, 2 ADDs
+        double f_alpha_beta_neg_x = r_ab_x * scalar_mult_r_ab + r_ab2_x * scalar_mult_r_ab2;
+        double f_alpha_beta_neg_y = r_ab_y * scalar_mult_r_ab + r_ab2_y * scalar_mult_r_ab2;
+
+        double w_val;
+
+        // # 3 MULTs, 1 ADD
+        double a = (p.e_x * f_alpha_beta_neg_x + p.e_y * f_alpha_beta_neg_y) * cos_phi_inv;
+
+        // # 2 MULTs, 1 ADD
+        double b_sq = f_alpha_beta_neg_x * f_alpha_beta_neg_x + f_alpha_beta_neg_y * f_alpha_beta_neg_y;
+
+        w_val = dt_neg;
+
+        // # 1 MULT
+        if (a >= 0 && (a * a > b_sq)) {
+          w_val = c_neg_times_dt;
+        }
+
+        // # 2 MULTs, 2 ADDs
+        f_alpha_dt_x = f_alpha_dt_x + f_alpha_beta_neg_x * w_val;
+        f_alpha_dt_y = f_alpha_dt_y + f_alpha_beta_neg_y * w_val;
+      }
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
+        double v1_x = p.curr_pos_x - board.obstacles[k].start.x;
+        double v1_y = p.curr_pos_y - board.obstacles[k].start.y;
+
+        // # 3 MULTs, 1 ADD
+        double projection = (v1_x * board.obstacles[k].obstacle_v.x + v1_y * board.obstacles[k].obstacle_v.y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0)
+          projection = 0.0;
+        else if (projection > 1.0)
+          projection = 1.0;
+
+        // # 4 ADDs, 2 MULTs
+        double r_aB_x = p.curr_pos_x - (board.obstacles[k].start.x + board.obstacles[k].obstacle_v.x * projection);
+        double r_aB_y = p.curr_pos_y - (board.obstacles[k].start.y + board.obstacles[k].obstacle_v.y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        double r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+
+        // # 4 MULTs, 1 EXP, 1 DIV
+        double scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+        double f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        double f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        double w_val;
+
+        // # 3 MULTs, 1 ADD
+        double a = (p.e_x * f_alpha_B_neg_x + p.e_y * f_alpha_B_neg_y) * cos_phi_inv;
+
+        // # 2 MULTS, 1 ADD
+        double b_sq = f_alpha_B_neg_x * f_alpha_B_neg_x + f_alpha_B_neg_y * f_alpha_B_neg_y;
+
+        w_val = dt_neg;
+
+        // # 1 MULT
+        if (a >= 0 && (a * a > b_sq)) {
+          w_val = c_neg_times_dt;
+        }
+
+        // # 2 MULTs
+        double f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        double f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        // # 2 ADDs
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      // # 2 ADDs
+      p.w_x += f_alpha_dt_x;
+      p.w_y += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+
+      // # 1 SQRT, 2 MULTs, 1 ADD
+      double w_norm = sqrt(p.w_x * p.w_x + p.w_y * p.w_y);
+
+      // # 1 DIV
+      double correction_factor = min(1.0, p.v_max / w_norm);
+
+      // # 2 MULTs
+      p.curr_vel_x = p.w_x * correction_factor;
+      p.curr_vel_y = p.w_y * correction_factor;
+
+      // # 2 MULTs, 2 ADDs
+      p.curr_pos_x = p.curr_pos_x + p.curr_vel_x * dt;
+      p.curr_pos_y = p.curr_pos_y + p.curr_vel_y * dt;
+
+      if (p.color == 0 && p.curr_pos_x > BOARD_LENGTH_MINUS_ONE_POINT_FIVE) {
+        p.curr_pos_x = 1.0;
+        p.destination_y = p.curr_pos_y;
+        p.w_x = p.desired_v0;
+        p.w_y = 0.0;
+        p.curr_vel_x = p.w_x;
+        p.curr_vel_y = p.w_y;
+      }
+      if (p.color == 1 && p.curr_pos_x < 1.5) {
+        p.curr_pos_x = BOARD_LENGTH_MINUS_ONE;
+        p.destination_y = p.curr_pos_y;
+        p.w_x = -p.desired_v0;
+        p.w_y = 0.0;
+        p.curr_vel_x = p.w_x;
+        p.curr_vel_y = p.w_y;
+      }
+
+      p.curr_pos_y = clamp(p.curr_pos_y, WALL_EPS, BOARD_WIDTH_MINUS_WALL_EPS);
+
+      // # 2 ADDs
+      double e_x = p.destination_x - p.curr_pos_x;
+      double e_y = p.destination_y - p.curr_pos_y;
+
+      // # 1 DIV, 1 SQRT, 2 MULTs, 1 ADD
+      double e_norm_inv = 1.0 / sqrt(e_x * e_x + e_y * e_y);
+
+      // # 2 MULTs
+      p.e_x = e_x * e_norm_inv;
+      p.e_y = e_y * e_norm_inv;
+
+      // # 2 MULTs, 1 ADD
+      double velocity_norm_sq = p.curr_vel_x * p.curr_vel_x + p.curr_vel_y * p.curr_vel_y;
+
+      // # 1 SQRT
+      double velocity_norm = sqrt(velocity_norm_sq);
+
+      // # 5 MULTs
+      p.s_b_sq = velocity_norm_sq * delta_t_sq;
+      p.s_b_e_x = p.e_x * (velocity_norm * delta_t);
+      p.s_b_e_y = p.e_y * (velocity_norm * delta_t);
+
+#if !BENCHMARK
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, p.curr_pos_x, p.curr_pos_y, p.e_x, p.e_y, p.color);
+      }
+#endif
+    }
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp b/variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp
new file mode 100644
index 0000000..cb3d8e3
--- /dev/null
+++ b/variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_expfix.cpp
@@ -0,0 +1,483 @@
+// # Changes:
+// # - Ported AVX2 to AVX-512
+// # - Replaced Obstacle with the version that contains no Vector2d
+// # - Implemented Fast Inverse Square Root
+// # Predecessor: variant_inlining_7_SoA_AVX2_polyexp.cpp
+
+#include "../headers/board_reduced.hpp"
+#include "../headers/constants.hpp"
+#include "../headers/pedestrian_reduced.hpp"
+
+#include <immintrin.h>
+
+#include <fstream>
+#include <random>
+#include <vector>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0
+#define BOARD_LENGTH 50.0
+#define WALL_EPS 1e-3
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+alignas(64) vector<double> pos_x, pos_y;
+alignas(64) vector<double> sb_sq;
+alignas(64) vector<double> sb_e_x, sb_e_y;
+alignas(64) vector<double> e_x, e_y;
+alignas(64) vector<double> curr_vel_x, curr_vel_y;
+alignas(64) vector<double> w_x, w_y;
+alignas(64) vector<double> desired_v0;
+alignas(64) vector<double> v_max;
+alignas(64) vector<double> destination_x, destination_y;
+alignas(64) vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<double> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<double> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<double> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<double> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  const int n_p_pad = (n_p + 7) & ~7;
+
+  pos_x.resize(n_p_pad);
+  pos_y.resize(n_p_pad);
+  sb_sq.resize(n_p_pad);
+  sb_e_x.resize(n_p_pad);
+  sb_e_y.resize(n_p_pad);
+  e_x.resize(n_p_pad);
+  e_y.resize(n_p_pad);
+  curr_vel_x.resize(n_p_pad);
+  curr_vel_y.resize(n_p_pad);
+  w_x.resize(n_p_pad);
+  w_y.resize(n_p_pad);
+  desired_v0.resize(n_p_pad);
+  v_max.resize(n_p_pad);
+  destination_x.resize(n_p_pad);
+  destination_y.resize(n_p_pad);
+  color.resize(n_p_pad);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    double x = x_gen_left(gen);
+    double y = y_gen(gen);
+
+    double dest_x = BOARD_LENGTH + 30.0;
+    double dest_y = y;
+
+    double init_e_x = dest_x - x;
+    double init_e_y = dest_y - y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1);
+
+    double v_m = 1.3 * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0;
+    w_y[i] = 0.0;
+    curr_vel_x[i] = 0.0;
+    curr_vel_y[i] = 0.0;
+    sb_sq[i] = 0.0;
+    sb_e_x[i] = 0.0;
+    sb_e_y[i] = 0.0;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    double x = x_gen_right(gen);
+    double y = y_gen(gen);
+
+    double dest_x = -30;
+    double dest_y = y;
+
+    double init_e_x = dest_x - x;
+    double init_e_y = dest_y - y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1);
+
+    double v_m = 1.3 * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0;
+    w_y[i] = 0.0;
+    curr_vel_x[i] = 0.0;
+    curr_vel_y[i] = 0.0;
+    sb_sq[i] = 0.0;
+    sb_e_x[i] = 0.0;
+    sb_e_y[i] = 0.0;
+  }
+
+  for (int j = n_p; j < n_p_pad; j++) {
+    pos_x[j] = 1e9;
+    pos_y[j] = 1e9;
+    sb_sq[j] = 0.0;
+    sb_e_x[j] = 0.0;
+    sb_e_y[j] = 0.0;
+  }
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+// # 128 MULTs, 112 ADDs
+static inline __m512d _avx512_exp_pd(__m512d x) {
+
+  const __m512d log2e = _mm512_set1_pd(1.4426950408889634);
+
+  // # 8 MULTs
+  __m512d nf = _mm512_roundscale_pd(_mm512_mul_pd(x, log2e), _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+  const __m512d ln2 = _mm512_set1_pd(0.6931471805599453);
+
+  // # 8 MULTs, 8 ADDs
+  __m512d r = _mm512_fnmadd_pd(nf, ln2, x);
+
+  const __m512d c13 = _mm512_set1_pd(c13_scalar);
+  const __m512d c12 = _mm512_set1_pd(c12_scalar);
+  const __m512d c11 = _mm512_set1_pd(c11_scalar);
+  const __m512d c10 = _mm512_set1_pd(c10_scalar);
+  const __m512d c9 = _mm512_set1_pd(c9_scalar);
+  const __m512d c8 = _mm512_set1_pd(c8_scalar);
+  const __m512d c7 = _mm512_set1_pd(c7_scalar);
+  const __m512d c6 = _mm512_set1_pd(c6_scalar);
+  const __m512d c5 = _mm512_set1_pd(c5_scalar);
+  const __m512d c4 = _mm512_set1_pd(c4_scalar);
+  const __m512d c3 = _mm512_set1_pd(c3_scalar);
+  const __m512d c2 = _mm512_set1_pd(c2_scalar);
+  const __m512d c1 = _mm512_set1_pd(c1_scalar);
+  const __m512d c0 = _mm512_set1_pd(c0_scalar);
+
+  __m512d p = c13;
+
+  // # 104 MULTs, 104 ADDs
+  p = _mm512_fmadd_pd(p, r, c12);
+  p = _mm512_fmadd_pd(p, r, c11);
+  p = _mm512_fmadd_pd(p, r, c10);
+  p = _mm512_fmadd_pd(p, r, c9);
+  p = _mm512_fmadd_pd(p, r, c8);
+  p = _mm512_fmadd_pd(p, r, c7);
+  p = _mm512_fmadd_pd(p, r, c6);
+  p = _mm512_fmadd_pd(p, r, c5);
+  p = _mm512_fmadd_pd(p, r, c4);
+  p = _mm512_fmadd_pd(p, r, c3);
+  p = _mm512_fmadd_pd(p, r, c2);
+  p = _mm512_fmadd_pd(p, r, c1);
+  p = _mm512_fmadd_pd(p, r, c0);
+
+  __m256d nf_lo = _mm512_castpd512_pd256(nf);
+  __m256d nf_hi = _mm512_extractf64x4_pd(nf, 1);
+
+  __m128i ni32_lo = _mm256_cvtpd_epi32(nf_lo);
+  __m128i ni32_hi = _mm256_cvtpd_epi32(nf_hi);
+
+  __m256i ni64_lo = _mm256_cvtepi32_epi64(ni32_lo);
+  __m256i ni64_hi = _mm256_cvtepi32_epi64(ni32_hi);
+
+  __m512i ni64 = _mm512_inserti64x4(_mm512_castsi256_si512(ni64_lo), ni64_hi, 1);
+
+  __m512i biased = _mm512_add_epi64(ni64, _mm512_set1_epi64(1023));
+
+  __m512i scaled = _mm512_slli_epi64(biased, 52);
+
+  __m512d pow2n = _mm512_castsi512_pd(scaled);
+
+  // # 8 MULTs
+  return _mm512_mul_pd(p, pow2n);
+}
+
+static inline double fast_inv_sqrt(double number) {
+  double y = number;
+
+  uint64_t i = bit_cast<uint64_t>(y);
+
+  i = 0x5fe6eb50c7b537a9ULL - (i >> 1);
+
+  y = bit_cast<double>(i);
+
+  double x2 = number * 0.5;
+  y = y * (1.5 - (x2 * y * y));
+  y = y * (1.5 - (x2 * y * y));
+  y = y * (1.5 - (x2 * y * y));
+
+  return y;
+}
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+
+  vector<tuple<int, int, double, double, double, double, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+
+    for (int i = 0; i < n_p; i++) {
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  const int n_p_pad = (n_p + 7) & ~7;
+
+  alignas(64) constexpr double lane_offset_arr[8] = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0};
+  const __m512d lane_off = _mm512_load_pd(lane_offset_arr);
+  const __m512d v_neg_v0_div_2s = _mm512_set1_pd(NEG_V0_AB_div_TWO_SIGMA);
+  const __m512d v_neg_2s_inv = _mm512_set1_pd(NEG_TWO_SIGMA_INV);
+  const __m512d v_cos_phi_inv = _mm512_set1_pd(cos_phi_inv);
+  const __m512d v_dt_neg = _mm512_set1_pd(dt_neg);
+  const __m512d v_c_neg_dt = _mm512_set1_pd(c_neg_times_dt);
+  const __m512d v_zero = _mm512_setzero_pd();
+
+  const __m512d v_np = _mm512_set1_pd(n_p);
+
+  // # ADDs: n_iter * n_p * 33 +
+  // #        n_iter * n_p * (n_p_pad / 8) * 232 +
+  // #        n_iter * n_p * board.n_obstacles * 10
+
+  // # MULTs: n_iter * n_p * 40 +
+  // #        n_iter * n_p * (n_p_pad / 8) * 320 +
+  // #        n_iter * n_p * board.n_obstacles * 13
+
+  // # DIVs: n_iter * n_p * (n_p_pad / 8) * 8 +
+  // #        n_iter * n_p * board.n_obstacles
+
+  // # SQRTs: n_iter * n_p +
+  // #        n_iter * n_p * (n_p_pad / 8) * 24 +
+  // #        n_iter * n_p * board.n_obstacles
+
+  // # EXPs: n_iter * n_p * board.n_obstacles
+
+  for (int t = 0; t < n_iter; t++) {
+
+    for (int i = 0; i < n_p; i++) {
+      double pos_x_i = pos_x[i];
+      double pos_y_i = pos_y[i];
+      double e_x_i = e_x[i];
+      double e_y_i = e_y[i];
+      double desired_v0_i = desired_v0[i];
+      double curr_vel_x_i = curr_vel_x[i];
+      double curr_vel_y_i = curr_vel_y[i];
+
+      double f_alpha_dt_x = (e_x_i * desired_v0_i - curr_vel_x_i) * dt_times_tau_alpha_inv;
+      double f_alpha_dt_y = (e_y_i * desired_v0_i - curr_vel_y_i) * dt_times_tau_alpha_inv;
+
+      const __m512d pix = _mm512_set1_pd(pos_x_i);
+      const __m512d piy = _mm512_set1_pd(pos_y_i);
+      const __m512d eix = _mm512_set1_pd(e_x_i);
+      const __m512d eiy = _mm512_set1_pd(e_y_i);
+      const __m512d v_i = _mm512_set1_pd((double)i);
+
+      __m512d acc_x = _mm512_setzero_pd();
+      __m512d acc_y = _mm512_setzero_pd();
+
+      for (int j = 0; j < n_p_pad; j += 8) {
+        __m512d j_idx = _mm512_add_pd(_mm512_set1_pd((double)j), lane_off);
+
+        __mmask8 m_neq_i = _mm512_cmp_pd_mask(j_idx, v_i, _CMP_NEQ_OQ);
+        __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_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);
+        __m512d r_ab_n2 = _mm512_fmadd_pd(r_ab_x, r_ab_x, _mm512_mul_pd(r_ab_y, r_ab_y));
+        __m512d r_ab_n = _mm512_sqrt_pd(r_ab_n2);
+
+        __m512d r_ab2_x = _mm512_sub_pd(r_ab_x, sbe_jx);
+        __m512d r_ab2_y = _mm512_sub_pd(r_ab_y, sbe_jy);
+        __m512d r_ab2_n2 = _mm512_fmadd_pd(r_ab2_x, r_ab2_x, _mm512_mul_pd(r_ab2_y, r_ab2_y));
+        __m512d r_ab2_n = _mm512_sqrt_pd(r_ab2_n2);
+
+        __m512d r_sum = _mm512_add_pd(r_ab_n, r_ab2_n);
+        __m512d two_b_sq = _mm512_fmsub_pd(r_sum, r_sum, sb_jsq);
+        __m512d two_b = _mm512_sqrt_pd(two_b_sq);
+
+        __m512d e_arg = _mm512_mul_pd(two_b, v_neg_2s_inv);
+        __m512d e_val = _avx512_exp_pd(e_arg);
+
+        __m512d numer = _mm512_mul_pd(_mm512_mul_pd(r_sum, v_neg_v0_div_2s), e_val);
+        __m512d denom = _mm512_mul_pd(_mm512_mul_pd(two_b, r_ab_n), r_ab2_n);
+        __m512d denom_safe = _mm512_mask_blend_pd(active, _mm512_set1_pd(1.0), denom);
+        __m512d smc = _mm512_div_pd(numer, denom_safe);
+
+        __m512d sm_r_ab = _mm512_mul_pd(smc, r_ab2_n);
+        __m512d sm_r_ab2 = _mm512_mul_pd(smc, r_ab_n);
+
+        __m512d f_x = _mm512_fmadd_pd(r_ab2_x, sm_r_ab2, _mm512_mul_pd(r_ab_x, sm_r_ab));
+        __m512d f_y = _mm512_fmadd_pd(r_ab2_y, sm_r_ab2, _mm512_mul_pd(r_ab_y, sm_r_ab));
+
+        __m512d a = _mm512_mul_pd(_mm512_fmadd_pd(eix, f_x, _mm512_mul_pd(eiy, f_y)), v_cos_phi_inv);
+        __m512d a_sq = _mm512_mul_pd(a, a);
+        __m512d b_sq = _mm512_fmadd_pd(f_x, f_x, _mm512_mul_pd(f_y, f_y));
+
+        __mmask8 k_ge = _mm512_cmp_pd_mask(a, v_zero, _CMP_GE_OQ);
+        __mmask8 k_gt = _mm512_cmp_pd_mask(a_sq, b_sq, _CMP_GT_OQ);
+        __mmask8 k_cond = _mm512_kand(k_ge, k_gt);
+        __m512d w_val = _mm512_mask_blend_pd(k_cond, v_dt_neg, v_c_neg_dt);
+
+        __m512d f_x_safe = _mm512_maskz_mov_pd(active, f_x);
+        __m512d f_y_safe = _mm512_maskz_mov_pd(active, f_y);
+        acc_x = _mm512_fmadd_pd(f_x_safe, w_val, acc_x);
+        acc_y = _mm512_fmadd_pd(f_y_safe, w_val, acc_y);
+      }
+
+      f_alpha_dt_x += _mm512_reduce_add_pd(acc_x);
+      f_alpha_dt_y += _mm512_reduce_add_pd(acc_y);
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        double v1_x = pos_x_i - board.obstacles[k].start_x;
+        double v1_y = pos_y_i - board.obstacles[k].start_y;
+
+        double projection = (v1_x * board.obstacles[k].obstacle_v_x + v1_y * board.obstacles[k].obstacle_v_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0)
+          projection = 0.0;
+        else if (projection > 1.0)
+          projection = 1.0;
+
+        double r_aB_x = pos_x_i - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        double r_aB_y = pos_y_i - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        double r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+        double scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+        double f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        double f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        double w_val = dt_neg;
+        double e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -e_y_i : e_y_i);
+        if (e_dot_f < cos_phi)
+          w_val = c_neg_times_dt;
+
+        f_alpha_dt_x += f_alpha_B_neg_x * w_val;
+        f_alpha_dt_y += f_alpha_B_neg_y * w_val;
+      }
+
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p; i++) {
+      double w_norm_inv = fast_inv_sqrt(w_x[i] * w_x[i] + w_y[i] * w_y[i]);
+
+      double correction_factor = min(1.0, v_max[i] * w_norm_inv);
+
+      curr_vel_x[i] = w_x[i] * correction_factor;
+      curr_vel_y[i] = w_y[i] * correction_factor;
+
+      pos_x[i] = pos_x[i] + curr_vel_x[i] * dt;
+      pos_y[i] = pos_y[i] + curr_vel_y[i] * dt;
+
+      if (!color[i] && pos_x[i] > (BOARD_LENGTH - 1.5)) {
+        pos_x[i] = 1.0;
+        destination_y[i] = pos_y[i];
+        w_x[i] = desired_v0[i];
+        w_y[i] = 0.0;
+        curr_vel_x[i] = w_x[i];
+        curr_vel_y[i] = w_y[i];
+      }
+      if (color[i] && pos_x[i] < 1.5) {
+        pos_x[i] = BOARD_LENGTH - 1.0;
+        destination_y[i] = pos_y[i];
+        w_x[i] = -desired_v0[i];
+        w_y[i] = 0.0;
+        curr_vel_x[i] = w_x[i];
+        curr_vel_y[i] = w_y[i];
+      }
+
+      pos_y[i] = clamp(pos_y[i], WALL_EPS, BOARD_WIDTH_MINUS_WALL_EPS);
+
+      double ex = destination_x[i] - pos_x[i];
+      double ey = destination_y[i] - pos_y[i];
+
+      double e_norm_inv = fast_inv_sqrt(ex * ex + ey * ey);
+
+      e_x[i] = ex * e_norm_inv;
+      e_y[i] = ey * e_norm_inv;
+
+      double velocity_norm_sq = curr_vel_x[i] * curr_vel_x[i] + curr_vel_y[i] * curr_vel_y[i];
+      double sb_sq_i = velocity_norm_sq * delta_t_sq;
+      double velocity_norm_mul_delta_t = sqrt(sb_sq_i);
+
+      sb_sq[i] = sb_sq_i;
+
+      sb_e_x[i] = e_x[i] * velocity_norm_mul_delta_t;
+      sb_e_y[i] = e_y[i] * velocity_norm_mul_delta_t;
+
+#if !BENCHMARK
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+#endif
+    }
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix.cpp b/variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix.cpp
new file mode 100644
index 0000000..c621275
--- /dev/null
+++ b/variants/variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt_sr_expfix.cpp
@@ -0,0 +1,576 @@
+// # Changes: Performed Scalar Replacements
+// # Predecessor: variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt.cpp
+
+#include "../headers/board_reduced.hpp"
+#include "../headers/constants.hpp"
+#include "../headers/pedestrian_reduced.hpp"
+
+#include <immintrin.h>
+
+#include <fstream>
+#include <random>
+#include <vector>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0
+#define BOARD_LENGTH 50.0
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5
+#define WALL_EPS 1e-3
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+alignas(64) vector<double> pos_x, pos_y;
+alignas(64) vector<double> sb_sq;
+alignas(64) vector<double> sb_e_x, sb_e_y;
+alignas(64) vector<double> e_x, e_y;
+alignas(64) vector<double> curr_vel_x, curr_vel_y;
+alignas(64) vector<double> w_x, w_y;
+alignas(64) vector<double> desired_v0;
+alignas(64) vector<double> v_max;
+alignas(64) vector<double> destination_x, destination_y;
+alignas(64) vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<double> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<double> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<double> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<double> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  const int n_p_pad = (n_p + 7) & ~7;
+
+  pos_x.resize(n_p_pad);
+  pos_y.resize(n_p_pad);
+  sb_sq.resize(n_p_pad);
+  sb_e_x.resize(n_p_pad);
+  sb_e_y.resize(n_p_pad);
+  e_x.resize(n_p_pad);
+  e_y.resize(n_p_pad);
+  curr_vel_x.resize(n_p_pad);
+  curr_vel_y.resize(n_p_pad);
+  w_x.resize(n_p_pad);
+  w_y.resize(n_p_pad);
+  desired_v0.resize(n_p_pad);
+  v_max.resize(n_p_pad);
+  destination_x.resize(n_p_pad);
+  destination_y.resize(n_p_pad);
+  color.resize(n_p_pad);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    double x = x_gen_left(gen);
+    double y = y_gen(gen);
+
+    double dest_x = BOARD_LENGTH + 30.0;
+    double dest_y = y;
+
+    double init_e_x = dest_x - x;
+    double init_e_y = dest_y - y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1);
+
+    double v_m = 1.3 * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0;
+    w_y[i] = 0.0;
+    curr_vel_x[i] = 0.0;
+    curr_vel_y[i] = 0.0;
+    sb_sq[i] = 0.0;
+    sb_e_x[i] = 0.0;
+    sb_e_y[i] = 0.0;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    double x = x_gen_right(gen);
+    double y = y_gen(gen);
+
+    double dest_x = -30;
+    double dest_y = y;
+
+    double init_e_x = dest_x - x;
+    double init_e_y = dest_y - y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1);
+
+    double v_m = 1.3 * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0;
+    w_y[i] = 0.0;
+    curr_vel_x[i] = 0.0;
+    curr_vel_y[i] = 0.0;
+    sb_sq[i] = 0.0;
+    sb_e_x[i] = 0.0;
+    sb_e_y[i] = 0.0;
+  }
+
+  for (int j = n_p; j < n_p_pad; j++) {
+    pos_x[j] = 1e9;
+    pos_y[j] = 1e9;
+    sb_sq[j] = 0.0;
+    sb_e_x[j] = 0.0;
+    sb_e_y[j] = 0.0;
+  }
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+// # 128 MULTs, 112 ADDs
+static inline __m512d _avx512_exp_pd(__m512d x) {
+
+  const __m512d log2e = _mm512_set1_pd(1.4426950408889634);
+
+  // # 8 MULTs
+  __m512d nf = _mm512_roundscale_pd(_mm512_mul_pd(x, log2e), _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+  const __m512d ln2 = _mm512_set1_pd(0.6931471805599453);
+
+  // # 8 MULTs, 8 ADDs
+  __m512d r = _mm512_fnmadd_pd(nf, ln2, x);
+
+  const __m512d c13 = _mm512_set1_pd(c13_scalar);
+  const __m512d c12 = _mm512_set1_pd(c12_scalar);
+  const __m512d c11 = _mm512_set1_pd(c11_scalar);
+  const __m512d c10 = _mm512_set1_pd(c10_scalar);
+  const __m512d c9 = _mm512_set1_pd(c9_scalar);
+  const __m512d c8 = _mm512_set1_pd(c8_scalar);
+  const __m512d c7 = _mm512_set1_pd(c7_scalar);
+  const __m512d c6 = _mm512_set1_pd(c6_scalar);
+  const __m512d c5 = _mm512_set1_pd(c5_scalar);
+  const __m512d c4 = _mm512_set1_pd(c4_scalar);
+  const __m512d c3 = _mm512_set1_pd(c3_scalar);
+  const __m512d c2 = _mm512_set1_pd(c2_scalar);
+  const __m512d c1 = _mm512_set1_pd(c1_scalar);
+  const __m512d c0 = _mm512_set1_pd(c0_scalar);
+
+  __m512d p = c13;
+
+  // # 104 MULTs, 104 ADDs
+  p = _mm512_fmadd_pd(p, r, c12);
+  p = _mm512_fmadd_pd(p, r, c11);
+  p = _mm512_fmadd_pd(p, r, c10);
+  p = _mm512_fmadd_pd(p, r, c9);
+  p = _mm512_fmadd_pd(p, r, c8);
+  p = _mm512_fmadd_pd(p, r, c7);
+  p = _mm512_fmadd_pd(p, r, c6);
+  p = _mm512_fmadd_pd(p, r, c5);
+  p = _mm512_fmadd_pd(p, r, c4);
+  p = _mm512_fmadd_pd(p, r, c3);
+  p = _mm512_fmadd_pd(p, r, c2);
+  p = _mm512_fmadd_pd(p, r, c1);
+  p = _mm512_fmadd_pd(p, r, c0);
+
+  __m256d nf_lo = _mm512_castpd512_pd256(nf);
+  __m256d nf_hi = _mm512_extractf64x4_pd(nf, 1);
+
+  __m128i ni32_lo = _mm256_cvtpd_epi32(nf_lo);
+  __m128i ni32_hi = _mm256_cvtpd_epi32(nf_hi);
+
+  __m256i ni64_lo = _mm256_cvtepi32_epi64(ni32_lo);
+  __m256i ni64_hi = _mm256_cvtepi32_epi64(ni32_hi);
+
+  __m512i ni64 = _mm512_inserti64x4(_mm512_castsi256_si512(ni64_lo), ni64_hi, 1);
+
+  __m512i biased = _mm512_add_epi64(ni64, _mm512_set1_epi64(1023));
+
+  __m512i scaled = _mm512_slli_epi64(biased, 52);
+
+  __m512d pow2n = _mm512_castsi512_pd(scaled);
+
+  // # 8 MULTs
+  return _mm512_mul_pd(p, pow2n);
+}
+
+// # 10 MULTs, 3 ADDs
+static inline double fast_inv_sqrt(double number) {
+  double y = number;
+
+  uint64_t i = bit_cast<uint64_t>(y);
+
+  i = 0x5fe6eb50c7b537a9ULL - (i >> 1);
+
+  y = bit_cast<double>(i);
+
+  double x2 = number * 0.5;
+  y = y * (1.5 - (x2 * y * y));
+  y = y * (1.5 - (x2 * y * y));
+  y = y * (1.5 - (x2 * y * y));
+
+  return y;
+}
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+
+  vector<tuple<int, int, double, double, double, double, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+
+    for (int i = 0; i < n_p; i++) {
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  const int n_p_pad = (n_p + 7) & ~7;
+
+  alignas(64) constexpr double lane_offset_arr[8] = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0};
+  const __m512d lane_off = _mm512_load_pd(lane_offset_arr);
+  const __m512d v_neg_v0_div_2s = _mm512_set1_pd(NEG_V0_AB_div_TWO_SIGMA);
+  const __m512d v_neg_2s_inv = _mm512_set1_pd(NEG_TWO_SIGMA_INV);
+  const __m512d v_cos_phi_inv = _mm512_set1_pd(cos_phi_inv);
+  const __m512d v_dt_neg = _mm512_set1_pd(dt_neg);
+  const __m512d v_c_neg_dt = _mm512_set1_pd(c_neg_times_dt);
+  const __m512d v_zero = _mm512_setzero_pd();
+
+  const __m512d v_np = _mm512_set1_pd(n_p);
+
+  // # ADDs: n_iter * n_p * 33 +
+  // #        n_iter * n_p * (n_p_pad / 8) * 232 +
+  // #        n_iter * n_p * board.n_obstacles * 10
+
+  // # MULTs: n_iter * n_p * 40 +
+  // #        n_iter * n_p * (n_p_pad / 8) * 320 +
+  // #        n_iter * n_p * board.n_obstacles * 13
+
+  // # DIVs: n_iter * n_p * (n_p_pad / 8) * 8 +
+  // #        n_iter * n_p * board.n_obstacles
+
+  // # SQRTs: n_iter * n_p +
+  // #        n_iter * n_p * (n_p_pad / 8) * 24 +
+  // #        n_iter * n_p * board.n_obstacles
+
+  // # EXPs: n_iter * n_p * board.n_obstacles
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+
+      double pos_x_i = pos_x[i];
+      double pos_y_i = pos_y[i];
+      double e_x_i = e_x[i];
+      double e_y_i = e_y[i];
+      double desired_v0_i = desired_v0[i];
+      double curr_vel_x_i = curr_vel_x[i];
+      double curr_vel_y_i = curr_vel_y[i];
+
+      // # 4 MULTs, 2 ADDs
+      double f_alpha_dt_x = (e_x_i * desired_v0_i - curr_vel_x_i) * dt_times_tau_alpha_inv;
+      double f_alpha_dt_y = (e_y_i * desired_v0_i - curr_vel_y_i) * dt_times_tau_alpha_inv;
+
+      const __m512d pix = _mm512_set1_pd(pos_x_i);
+      const __m512d piy = _mm512_set1_pd(pos_y_i);
+      const __m512d eix = _mm512_set1_pd(e_x_i);
+      const __m512d eiy = _mm512_set1_pd(e_y_i);
+      const __m512d v_i = _mm512_set1_pd((double)i);
+
+      __m512d acc_x = _mm512_setzero_pd();
+      __m512d acc_y = _mm512_setzero_pd();
+
+      for (int j = 0; j < n_p_pad; j += 8) {
+
+        // # 8 ADDs
+        __m512d j_idx = _mm512_add_pd(_mm512_set1_pd((double)j), lane_off);
+
+        __mmask8 m_neq_i = _mm512_cmp_pd_mask(j_idx, v_i, _CMP_NEQ_OQ);
+        __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_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]);
+
+        // # 16 ADDs
+        __m512d r_ab_x = _mm512_sub_pd(pix, pjx);
+        __m512d r_ab_y = _mm512_sub_pd(piy, pjy);
+
+        // # 16 MULTs, 8 ADDs
+        __m512d r_ab_n2 = _mm512_fmadd_pd(r_ab_x, r_ab_x, _mm512_mul_pd(r_ab_y, r_ab_y));
+
+        // # 8 SQRTs
+        __m512d r_ab_n = _mm512_sqrt_pd(r_ab_n2);
+
+        // # 16 ADDs
+        __m512d r_ab2_x = _mm512_sub_pd(r_ab_x, sbe_jx);
+        __m512d r_ab2_y = _mm512_sub_pd(r_ab_y, sbe_jy);
+
+        // # 16 MULTs, 8 ADDs
+        __m512d r_ab2_n2 = _mm512_fmadd_pd(r_ab2_x, r_ab2_x, _mm512_mul_pd(r_ab2_y, r_ab2_y));
+
+        // # 8 SQRTs
+        __m512d r_ab2_n = _mm512_sqrt_pd(r_ab2_n2);
+
+        // # 8 ADDs
+        __m512d r_sum = _mm512_add_pd(r_ab_n, r_ab2_n);
+
+        // # 8 MULTs, 8 ADDs
+        __m512d two_b_sq = _mm512_fmsub_pd(r_sum, r_sum, sb_jsq);
+
+        // # 8 SQRTs
+        __m512d two_b = _mm512_sqrt_pd(two_b_sq);
+
+        // # 8 MULTs
+        __m512d e_arg = _mm512_mul_pd(two_b, v_neg_2s_inv);
+
+        // # 128 MULTs, 112 ADDs
+        __m512d e_val = _avx512_exp_pd(e_arg);
+
+        // # 32 MULTs
+        __m512d numer = _mm512_mul_pd(_mm512_mul_pd(r_sum, v_neg_v0_div_2s), e_val);
+        __m512d denom = _mm512_mul_pd(_mm512_mul_pd(two_b, r_ab_n), r_ab2_n);
+
+        __m512d denom_safe = _mm512_mask_blend_pd(active, _mm512_set1_pd(1.0), denom);
+
+        // # 8 DIVs
+        __m512d smc = _mm512_div_pd(numer, denom_safe);
+
+        // # 16 MULTs
+        __m512d sm_r_ab = _mm512_mul_pd(smc, r_ab2_n);
+        __m512d sm_r_ab2 = _mm512_mul_pd(smc, r_ab_n);
+
+        // # 32 MULTs, 16 ADDs
+        __m512d f_x = _mm512_fmadd_pd(r_ab2_x, sm_r_ab2, _mm512_mul_pd(r_ab_x, sm_r_ab));
+        __m512d f_y = _mm512_fmadd_pd(r_ab2_y, sm_r_ab2, _mm512_mul_pd(r_ab_y, sm_r_ab));
+
+        // # 24 MULTs, 8 ADDs
+        __m512d a = _mm512_mul_pd(_mm512_fmadd_pd(eix, f_x, _mm512_mul_pd(eiy, f_y)), v_cos_phi_inv);
+
+        // # 8 MULTs
+        __m512d a_sq = _mm512_mul_pd(a, a);
+
+        // # 16 MULTs, 8 ADDs
+        __m512d b_sq = _mm512_fmadd_pd(f_x, f_x, _mm512_mul_pd(f_y, f_y));
+
+        __mmask8 k_ge = _mm512_cmp_pd_mask(a, v_zero, _CMP_GE_OQ);
+        __mmask8 k_gt = _mm512_cmp_pd_mask(a_sq, b_sq, _CMP_GT_OQ);
+        __mmask8 k_cond = _mm512_kand(k_ge, k_gt);
+        __m512d w_val = _mm512_mask_blend_pd(k_cond, v_dt_neg, v_c_neg_dt);
+
+        __m512d f_x_safe = _mm512_maskz_mov_pd(active, f_x);
+        __m512d f_y_safe = _mm512_maskz_mov_pd(active, f_y);
+
+        // # 16 MULTs, 16 ADDs
+        acc_x = _mm512_fmadd_pd(f_x_safe, w_val, acc_x);
+        acc_y = _mm512_fmadd_pd(f_y_safe, w_val, acc_y);
+      }
+
+      // # 16 ADDs
+      f_alpha_dt_x += _mm512_reduce_add_pd(acc_x);
+      f_alpha_dt_y += _mm512_reduce_add_pd(acc_y);
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        double obs_start_x = board.obstacles[k].start_x;
+        double obs_start_y = board.obstacles[k].start_y;
+        double obs_v_x = board.obstacles[k].obstacle_v_x;
+        double obs_v_y = board.obstacles[k].obstacle_v_y;
+        double obs_length_sq_inv = board.obstacles[k].length_sq_inv;
+
+        // # 2 ADDs
+        double v1_x = pos_x_i - obs_start_x;
+        double v1_y = pos_y_i - obs_start_y;
+
+        // # 3 MULTs, 1 ADD
+        double projection = (v1_x * obs_v_x + v1_y * obs_v_y) * obs_length_sq_inv;
+
+        if (projection < 0.0)
+          projection = 0.0;
+        else if (projection > 1.0)
+          projection = 1.0;
+
+        // # 2 MULTs, 4 ADDs
+        double r_aB_x = pos_x_i - (obs_start_x + obs_v_x * projection);
+        double r_aB_y = pos_y_i - (obs_start_y + obs_v_y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        double r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+
+        // # 2 MULTs, 1 EXP, 1 DIV
+        double scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+
+        // # 2 MULTs
+        double f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        double f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        double w_val = dt_neg;
+        double e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -e_y_i : e_y_i);
+        if (e_dot_f < cos_phi)
+          w_val = c_neg_times_dt;
+
+        // # 2 MULTs, 2 ADDs
+        f_alpha_dt_x += f_alpha_B_neg_x * w_val;
+        f_alpha_dt_y += f_alpha_B_neg_y * w_val;
+      }
+
+      // # 2 ADDs
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p; i++) {
+
+      double color_i = color[i];
+      double curr_vel_x_i = curr_vel_x[i];
+      double curr_vel_y_i = curr_vel_y[i];
+      double w_x_i = w_x[i];
+      double w_y_i = w_y[i];
+      double pos_x_i = pos_x[i];
+      double pos_y_i = pos_y[i];
+      double e_x_i = e_x[i];
+      double e_y_i = e_y[i];
+      double v_max_i = v_max[i];
+      double desired_v0_i = desired_v0[i];
+      double destination_x_i = destination_x[i];
+      double destination_y_i = destination_y[i];
+
+      // # 12 MULTs, 4 ADDs
+      double w_norm_inv = fast_inv_sqrt(w_x_i * w_x_i + w_y_i * w_y_i);
+
+      // # 1 MULT
+      double correction_factor = min(1.0, v_max_i * w_norm_inv);
+
+      // # 2 MULTs
+      curr_vel_x_i = w_x_i * correction_factor;
+      curr_vel_y_i = w_y_i * correction_factor;
+
+      // # 2 MULTs, 2 ADDs
+      pos_x_i = pos_x_i + curr_vel_x_i * dt;
+      pos_y_i = pos_y_i + curr_vel_y_i * dt;
+
+      if (!color_i && pos_x_i > (BOARD_LENGTH_MINUS_ONE_POINT_FIVE)) {
+        pos_x_i = 1.0;
+        destination_y_i = pos_y_i;
+        w_x_i = desired_v0_i;
+        w_y_i = 0.0;
+        curr_vel_x_i = w_x_i;
+        curr_vel_y_i = w_y_i;
+      }
+      if (color_i && pos_x_i < 1.5) {
+        pos_x_i = BOARD_LENGTH_MINUS_ONE;
+        destination_y_i = pos_y_i;
+        w_x_i = -desired_v0_i;
+        w_y_i = 0.0;
+        curr_vel_x_i = w_x_i;
+        curr_vel_y_i = w_y_i;
+      }
+
+      pos_y_i = clamp(pos_y_i, WALL_EPS, BOARD_WIDTH_MINUS_WALL_EPS);
+
+      // # 2 ADDs
+      double ex = destination_x_i - pos_x_i;
+      double ey = destination_y_i - pos_y_i;
+
+      // # 12 MULTs, 4 ADDs
+      double e_norm_inv = fast_inv_sqrt(ex * ex + ey * ey);
+
+      // # 2 MULTs
+      e_x_i = ex * e_norm_inv;
+      e_y_i = ey * e_norm_inv;
+
+      // # 2 MULTs, 1 ADD
+      double velocity_norm_sq = curr_vel_x_i * curr_vel_x_i + curr_vel_y_i * curr_vel_y_i;
+
+      // # 1 MULT
+      double sb_sq_i = velocity_norm_sq * delta_t_sq;
+
+      // # 1 SQRT
+      double velocity_norm_mul_delta_t = sqrt(sb_sq_i);
+
+      curr_vel_x[i] = curr_vel_x_i;
+      curr_vel_y[i] = curr_vel_y_i;
+
+      w_x[i] = w_x_i;
+      w_y[i] = w_y_i;
+
+      pos_x[i] = pos_x_i;
+      pos_y[i] = pos_y_i;
+
+      e_x[i] = e_x_i;
+      e_y[i] = e_y_i;
+
+      destination_y[i] = destination_y_i;
+
+      sb_sq[i] = sb_sq_i;
+
+      // # 2 MULTs
+      sb_e_x[i] = e_x_i * velocity_norm_mul_delta_t;
+      sb_e_y[i] = e_y_i * velocity_norm_mul_delta_t;
+
+#if !BENCHMARK
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+#endif
+    }
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix.cpp b/variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix.cpp
new file mode 100644
index 0000000..7b51d9a
--- /dev/null
+++ b/variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_expfix.cpp
@@ -0,0 +1,563 @@
+// # Changes: Vectorised second pass with AVX-512
+// # Predecessor: variant_inlining_8_SoA_AVX512_polyexp_fastinvsqrt.cpp
+
+#include "../headers/board_reduced.hpp"
+#include "../headers/constants.hpp"
+#include "../headers/pedestrian_reduced.hpp"
+
+#include <immintrin.h>
+
+#include <fstream>
+#include <random>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0
+#define BOARD_LENGTH 50.0
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5
+#define WALL_EPS 1e-3
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+alignas(64) vector<double> pos_x, pos_y;
+alignas(64) vector<double> sb_sq;
+alignas(64) vector<double> sb_e_x, sb_e_y;
+alignas(64) vector<double> e_x, e_y;
+alignas(64) vector<double> curr_vel_x, curr_vel_y;
+alignas(64) vector<double> w_x, w_y;
+alignas(64) vector<double> desired_v0;
+alignas(64) vector<double> v_max;
+alignas(64) vector<double> destination_x, destination_y;
+alignas(64) vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<double> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<double> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<double> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<double> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  const int n_p_pad = (n_p + 7) & ~7;
+
+  pos_x.resize(n_p_pad);
+  pos_y.resize(n_p_pad);
+  sb_sq.resize(n_p_pad);
+  sb_e_x.resize(n_p_pad);
+  sb_e_y.resize(n_p_pad);
+  e_x.resize(n_p_pad);
+  e_y.resize(n_p_pad);
+  curr_vel_x.resize(n_p_pad);
+  curr_vel_y.resize(n_p_pad);
+  w_x.resize(n_p_pad);
+  w_y.resize(n_p_pad);
+  desired_v0.resize(n_p_pad);
+  v_max.resize(n_p_pad);
+  destination_x.resize(n_p_pad);
+  destination_y.resize(n_p_pad);
+  color.resize(n_p_pad);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    double x = x_gen_left(gen);
+    double y = y_gen(gen);
+
+    double dest_x = BOARD_LENGTH + 30.0;
+    double dest_y = y;
+
+    double init_e_x = dest_x - x;
+    double init_e_y = dest_y - y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1);
+
+    double v_m = 1.3 * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0;
+    w_y[i] = 0.0;
+    curr_vel_x[i] = 0.0;
+    curr_vel_y[i] = 0.0;
+    sb_sq[i] = 0.0;
+    sb_e_x[i] = 0.0;
+    sb_e_y[i] = 0.0;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    double x = x_gen_right(gen);
+    double y = y_gen(gen);
+
+    double dest_x = -30;
+    double dest_y = y;
+
+    double init_e_x = dest_x - x;
+    double init_e_y = dest_y - y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1);
+
+    double v_m = 1.3 * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0;
+    w_y[i] = 0.0;
+    curr_vel_x[i] = 0.0;
+    curr_vel_y[i] = 0.0;
+    sb_sq[i] = 0.0;
+    sb_e_x[i] = 0.0;
+    sb_e_y[i] = 0.0;
+  }
+
+  for (int j = n_p; j < n_p_pad; j++) {
+    pos_x[j] = 1e9;
+    pos_y[j] = 1e9;
+    sb_sq[j] = 0.0;
+    sb_e_x[j] = 0.0;
+    sb_e_y[j] = 0.0;
+  }
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+// # 128 MULTs, 112 ADDs
+static inline __m512d _avx512_exp_pd(__m512d x) {
+
+  const __m512d log2e = _mm512_set1_pd(1.4426950408889634);
+
+  // # 8 MULTs
+  __m512d nf = _mm512_roundscale_pd(_mm512_mul_pd(x, log2e), _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+  const __m512d ln2 = _mm512_set1_pd(0.6931471805599453);
+
+  // # 8 MULTs, 8 ADDs
+  __m512d r = _mm512_fnmadd_pd(nf, ln2, x);
+
+  const __m512d c13 = _mm512_set1_pd(c13_scalar);
+  const __m512d c12 = _mm512_set1_pd(c12_scalar);
+  const __m512d c11 = _mm512_set1_pd(c11_scalar);
+  const __m512d c10 = _mm512_set1_pd(c10_scalar);
+  const __m512d c9 = _mm512_set1_pd(c9_scalar);
+  const __m512d c8 = _mm512_set1_pd(c8_scalar);
+  const __m512d c7 = _mm512_set1_pd(c7_scalar);
+  const __m512d c6 = _mm512_set1_pd(c6_scalar);
+  const __m512d c5 = _mm512_set1_pd(c5_scalar);
+  const __m512d c4 = _mm512_set1_pd(c4_scalar);
+  const __m512d c3 = _mm512_set1_pd(c3_scalar);
+  const __m512d c2 = _mm512_set1_pd(c2_scalar);
+  const __m512d c1 = _mm512_set1_pd(c1_scalar);
+  const __m512d c0 = _mm512_set1_pd(c0_scalar);
+
+  __m512d p = c13;
+
+  // # 104 MULTs, 104 ADDs
+  p = _mm512_fmadd_pd(p, r, c12);
+  p = _mm512_fmadd_pd(p, r, c11);
+  p = _mm512_fmadd_pd(p, r, c10);
+  p = _mm512_fmadd_pd(p, r, c9);
+  p = _mm512_fmadd_pd(p, r, c8);
+  p = _mm512_fmadd_pd(p, r, c7);
+  p = _mm512_fmadd_pd(p, r, c6);
+  p = _mm512_fmadd_pd(p, r, c5);
+  p = _mm512_fmadd_pd(p, r, c4);
+  p = _mm512_fmadd_pd(p, r, c3);
+  p = _mm512_fmadd_pd(p, r, c2);
+  p = _mm512_fmadd_pd(p, r, c1);
+  p = _mm512_fmadd_pd(p, r, c0);
+
+  __m256d nf_lo = _mm512_castpd512_pd256(nf);
+  __m256d nf_hi = _mm512_extractf64x4_pd(nf, 1);
+
+  __m128i ni32_lo = _mm256_cvtpd_epi32(nf_lo);
+  __m128i ni32_hi = _mm256_cvtpd_epi32(nf_hi);
+
+  __m256i ni64_lo = _mm256_cvtepi32_epi64(ni32_lo);
+  __m256i ni64_hi = _mm256_cvtepi32_epi64(ni32_hi);
+
+  __m512i ni64 = _mm512_inserti64x4(_mm512_castsi256_si512(ni64_lo), ni64_hi, 1);
+
+  __m512i biased = _mm512_add_epi64(ni64, _mm512_set1_epi64(1023));
+
+  __m512i scaled = _mm512_slli_epi64(biased, 52);
+
+  __m512d pow2n = _mm512_castsi512_pd(scaled);
+
+  // # 8 MULTs
+  return _mm512_mul_pd(p, pow2n);
+}
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+
+  vector<tuple<int, int, double, double, double, double, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+
+    for (int i = 0; i < n_p; i++) {
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  const int n_p_pad = (n_p + 7) & ~7;
+
+  alignas(64) constexpr double lane_offset_arr[8] = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0};
+  const __m512d lane_off = _mm512_load_pd(lane_offset_arr);
+  const __m512d v_neg_v0_div_2s = _mm512_set1_pd(NEG_V0_AB_div_TWO_SIGMA);
+  const __m512d v_neg_2s_inv = _mm512_set1_pd(NEG_TWO_SIGMA_INV);
+  const __m512d v_cos_phi_inv = _mm512_set1_pd(cos_phi_inv);
+  const __m512d v_dt_neg = _mm512_set1_pd(dt_neg);
+  const __m512d v_c_neg_dt = _mm512_set1_pd(c_neg_times_dt);
+  const __m512d v_zero = _mm512_setzero_pd();
+
+  const __m512d v_np = _mm512_set1_pd(n_p);
+
+  // # ADDs: n_iter * n_p * 20 +
+  // #        n_iter * n_p * (n_p_pad / 8) * 232 +
+  // #        n_iter * n_p * board.n_obstacles * 10 +
+  // #        n_iter * (n_p_pad / 8) * 56
+
+  // # MULTs: n_iter * n_p * 4 +
+  // #        n_iter * n_p * (n_p_pad / 8) * 320 +
+  // #        n_iter * n_p * board.n_obstacles * 13 +
+  // #        n_iter * (n_p_pad / 8) * 120
+
+  // # DIVs: n_iter * n_p * (n_p_pad / 8) * 8 +
+  // #        n_iter * n_p * board.n_obstacles +
+  // #        n_iter * (n_p_pad / 8) * 16
+
+  // # SQRTs: n_iter * n_p * (n_p_pad / 8) * 24 +
+  // #        n_iter * n_p * board.n_obstacles +
+  // #        n_iter * (n_p_pad / 8) * 24
+
+  // # EXPs: n_iter * n_p * board.n_obstacles
+
+  for (int t = 0; t < n_iter; t++) {
+
+    for (int i = 0; i < n_p; i++) {
+      double pos_x_i = pos_x[i];
+      double pos_y_i = pos_y[i];
+      double e_x_i = e_x[i];
+      double e_y_i = e_y[i];
+      double desired_v0_i = desired_v0[i];
+      double curr_vel_x_i = curr_vel_x[i];
+      double curr_vel_y_i = curr_vel_y[i];
+
+      // # 4 MULTs, 2 ADDs
+      double f_alpha_dt_x = (e_x_i * desired_v0_i - curr_vel_x_i) * dt_times_tau_alpha_inv;
+      double f_alpha_dt_y = (e_y_i * desired_v0_i - curr_vel_y_i) * dt_times_tau_alpha_inv;
+
+      const __m512d pix = _mm512_set1_pd(pos_x_i);
+      const __m512d piy = _mm512_set1_pd(pos_y_i);
+      const __m512d eix = _mm512_set1_pd(e_x_i);
+      const __m512d eiy = _mm512_set1_pd(e_y_i);
+      const __m512d v_i = _mm512_set1_pd((double)i);
+
+      __m512d acc_x = _mm512_setzero_pd();
+      __m512d acc_y = _mm512_setzero_pd();
+
+      for (int j = 0; j < n_p_pad; j += 8) {
+
+        // # 8 ADDs
+        __m512d j_idx = _mm512_add_pd(_mm512_set1_pd((double)j), lane_off);
+
+        __mmask8 m_neq_i = _mm512_cmp_pd_mask(j_idx, v_i, _CMP_NEQ_OQ);
+        __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_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]);
+
+        // # 16 ADDs
+        __m512d r_ab_x = _mm512_sub_pd(pix, pjx);
+        __m512d r_ab_y = _mm512_sub_pd(piy, pjy);
+
+        // # 16 MULTs, 8 ADDs
+        __m512d r_ab_n2 = _mm512_fmadd_pd(r_ab_x, r_ab_x, _mm512_mul_pd(r_ab_y, r_ab_y));
+
+        // # 8 SQRTs
+        __m512d r_ab_n = _mm512_sqrt_pd(r_ab_n2);
+
+        // # 16 ADDs
+        __m512d r_ab2_x = _mm512_sub_pd(r_ab_x, sbe_jx);
+        __m512d r_ab2_y = _mm512_sub_pd(r_ab_y, sbe_jy);
+
+        // # 16 MULTs, 8 ADDs
+        __m512d r_ab2_n2 = _mm512_fmadd_pd(r_ab2_x, r_ab2_x, _mm512_mul_pd(r_ab2_y, r_ab2_y));
+
+        // # 8 SQRTs
+        __m512d r_ab2_n = _mm512_sqrt_pd(r_ab2_n2);
+
+        // # 8 ADDs
+        __m512d r_sum = _mm512_add_pd(r_ab_n, r_ab2_n);
+
+        // # 8 MULTs, 8 ADDs
+        __m512d two_b_sq = _mm512_fmsub_pd(r_sum, r_sum, sb_jsq);
+
+        // # 8 SQRTs
+        __m512d two_b = _mm512_sqrt_pd(two_b_sq);
+
+        // # 8 MULTs
+        __m512d e_arg = _mm512_mul_pd(two_b, v_neg_2s_inv);
+
+        // # 128 MULTs, 112 ADDs
+        __m512d e_val = _avx512_exp_pd(e_arg);
+
+        // # 32 MULTs
+        __m512d numer = _mm512_mul_pd(_mm512_mul_pd(r_sum, v_neg_v0_div_2s), e_val);
+        __m512d denom = _mm512_mul_pd(_mm512_mul_pd(two_b, r_ab_n), r_ab2_n);
+
+        // # 8 DIVs
+        __m512d smc = _mm512_div_pd(numer, denom);
+
+        // # 16 MULTs
+        __m512d sm_r_ab = _mm512_mul_pd(smc, r_ab2_n);
+        __m512d sm_r_ab2 = _mm512_mul_pd(smc, r_ab_n);
+
+        // # 32 MULTs, 16 ADDs
+        __m512d f_x = _mm512_fmadd_pd(r_ab2_x, sm_r_ab2, _mm512_mul_pd(r_ab_x, sm_r_ab));
+        __m512d f_y = _mm512_fmadd_pd(r_ab2_y, sm_r_ab2, _mm512_mul_pd(r_ab_y, sm_r_ab));
+
+        // # 24 MULTs, 8 ADDs
+        __m512d a = _mm512_mul_pd(_mm512_fmadd_pd(eix, f_x, _mm512_mul_pd(eiy, f_y)), v_cos_phi_inv);
+
+        // # 8 MULTs
+        __m512d a_sq = _mm512_mul_pd(a, a);
+
+        // # 16 MULTs, 8 ADDs
+        __m512d b_sq = _mm512_fmadd_pd(f_x, f_x, _mm512_mul_pd(f_y, f_y));
+
+        __mmask8 k_ge = _mm512_cmp_pd_mask(a, v_zero, _CMP_GE_OQ);
+        __mmask8 k_gt = _mm512_cmp_pd_mask(a_sq, b_sq, _CMP_GT_OQ);
+        __mmask8 k_cond = _mm512_kand(k_ge, k_gt);
+        __m512d w_val = _mm512_mask_blend_pd(k_cond, v_dt_neg, v_c_neg_dt);
+
+        __m512d f_x_safe = _mm512_maskz_mov_pd(active, f_x);
+        __m512d f_y_safe = _mm512_maskz_mov_pd(active, f_y);
+
+        // # 16 MULTs, 16 ADDs
+        acc_x = _mm512_fmadd_pd(f_x_safe, w_val, acc_x);
+        acc_y = _mm512_fmadd_pd(f_y_safe, w_val, acc_y);
+      }
+
+      // # 16 ADDS
+      f_alpha_dt_x += _mm512_reduce_add_pd(acc_x);
+      f_alpha_dt_y += _mm512_reduce_add_pd(acc_y);
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
+        double v1_x = pos_x_i - board.obstacles[k].start_x;
+        double v1_y = pos_y_i - board.obstacles[k].start_y;
+
+        // # 3 MULTs, 1 ADD
+        double projection = (v1_x * board.obstacles[k].obstacle_v_x + v1_y * board.obstacles[k].obstacle_v_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0)
+          projection = 0.0;
+        else if (projection > 1.0)
+          projection = 1.0;
+
+        // # 4 ADDs, 2 MULTs
+        double r_aB_x = pos_x_i - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        double r_aB_y = pos_y_i - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        double r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+
+        // # 2 MULTs, 1 EXP, 1 DIV
+        double scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+
+        // # 2 MULTs
+        double f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        double f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        double w_val = dt_neg;
+        double e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -e_y_i : e_y_i);
+        if (e_dot_f < cos_phi)
+          w_val = c_neg_times_dt;
+
+        // # 2 MULTs, 2 ADDs
+        f_alpha_dt_x += f_alpha_B_neg_x * w_val;
+        f_alpha_dt_y += f_alpha_B_neg_y * w_val;
+      }
+
+      // # 2 ADDs
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p_pad; i += 8) {
+
+      __m512d w_x_i_vec = _mm512_loadu_pd(&w_x[i]);
+      __m512d w_y_i_vec = _mm512_loadu_pd(&w_y[i]);
+      __m512d v_max_i_vec = _mm512_loadu_pd(&v_max[i]);
+      __m512d pos_x_i_vec = _mm512_loadu_pd(&pos_x[i]);
+      __m512d pos_y_i_vec = _mm512_loadu_pd(&pos_y[i]);
+      __m512d destination_y_i_vec = _mm512_loadu_pd(&destination_y[i]);
+      __m512d destination_x_i_vec = _mm512_loadu_pd(&destination_x[i]);
+      __m512d desired_v0_i_vec = _mm512_loadu_pd(&desired_v0[i]);
+
+      __m256i color_i_vec = _mm256_loadu_si256((__m256i *)&color[i]);
+
+      // # 16 MULTs, 8 ADDs
+      __m512d w_norm_sq = _mm512_fmadd_pd(w_x_i_vec, w_x_i_vec, _mm512_mul_pd(w_y_i_vec, w_y_i_vec));
+
+      // # 8 SQRTs
+      __m512d w_norm = _mm512_sqrt_pd(w_norm_sq);
+
+      // # 8 DIVs
+      __m512d correction_factor = _mm512_min_pd(_mm512_set1_pd(1.0), _mm512_div_pd(v_max_i_vec, w_norm));
+
+      // # 16 MULTs
+      __m512d curr_vel_x_i_vec = _mm512_mul_pd(w_x_i_vec, correction_factor);
+      __m512d curr_vel_y_i_vec = _mm512_mul_pd(w_y_i_vec, correction_factor);
+
+      __m512d dt_vec = _mm512_set1_pd(dt);
+
+      // # 16 MULTs, 16 ADDs
+      pos_x_i_vec = _mm512_fmadd_pd(curr_vel_x_i_vec, dt_vec, pos_x_i_vec);
+      pos_y_i_vec = _mm512_fmadd_pd(curr_vel_y_i_vec, dt_vec, pos_y_i_vec);
+
+      __mmask8 cond_1_left = _mm256_cmpeq_epi32_mask(color_i_vec, _mm256_setzero_si256());
+      __mmask8 cond_1_right = _mm512_cmp_pd_mask(pos_x_i_vec, _mm512_set1_pd(BOARD_LENGTH_MINUS_ONE_POINT_FIVE), _CMP_GT_OQ);
+      __mmask8 cond_1 = cond_1_left & cond_1_right;
+
+      pos_x_i_vec = _mm512_mask_blend_pd(cond_1, pos_x_i_vec, _mm512_set1_pd(1.0));
+      destination_y_i_vec = _mm512_mask_blend_pd(cond_1, destination_y_i_vec, pos_y_i_vec);
+      w_x_i_vec = _mm512_mask_blend_pd(cond_1, w_x_i_vec, desired_v0_i_vec);
+      w_y_i_vec = _mm512_mask_blend_pd(cond_1, w_y_i_vec, _mm512_setzero_pd());
+      curr_vel_x_i_vec = _mm512_mask_blend_pd(cond_1, curr_vel_x_i_vec, w_x_i_vec);
+      curr_vel_y_i_vec = _mm512_mask_blend_pd(cond_1, curr_vel_y_i_vec, w_y_i_vec);
+
+      __mmask8 cond_2_left = _mm256_cmpeq_epi32_mask(color_i_vec, _mm256_set1_epi32(1));
+      __mmask8 cond_2_right = _mm512_cmp_pd_mask(pos_x_i_vec, _mm512_set1_pd(1.5), _CMP_LT_OQ);
+      __mmask8 cond_2 = cond_2_left & cond_2_right;
+
+      pos_x_i_vec = _mm512_mask_blend_pd(cond_2, pos_x_i_vec, _mm512_set1_pd(BOARD_LENGTH - 1.0));
+      destination_y_i_vec = _mm512_mask_blend_pd(cond_2, destination_y_i_vec, pos_y_i_vec);
+      __m512d neg_desired_v0 = _mm512_xor_pd(desired_v0_i_vec, _mm512_set1_pd(-0.0));
+      w_x_i_vec = _mm512_mask_blend_pd(cond_2, w_x_i_vec, neg_desired_v0);
+      w_y_i_vec = _mm512_mask_blend_pd(cond_2, w_y_i_vec, _mm512_setzero_pd());
+      curr_vel_x_i_vec = _mm512_mask_blend_pd(cond_2, curr_vel_x_i_vec, w_x_i_vec);
+      curr_vel_y_i_vec = _mm512_mask_blend_pd(cond_2, curr_vel_y_i_vec, w_y_i_vec);
+
+      pos_y_i_vec = _mm512_max_pd(pos_y_i_vec, _mm512_set1_pd(WALL_EPS));
+      pos_y_i_vec = _mm512_min_pd(pos_y_i_vec, _mm512_set1_pd(BOARD_WIDTH_MINUS_WALL_EPS));
+
+      // # 16 ADDs
+      __m512d ex_vec = _mm512_sub_pd(destination_x_i_vec, pos_x_i_vec);
+      __m512d ey_vec = _mm512_sub_pd(destination_y_i_vec, pos_y_i_vec);
+
+      // # 16 MULTs, 8 ADDs
+      __m512d e_norm_sq = _mm512_fmadd_pd(ex_vec, ex_vec, _mm512_mul_pd(ey_vec, ey_vec));
+
+      // # 8 SQRTs
+      __m512d e_norm = _mm512_sqrt_pd(e_norm_sq);
+
+      // # 8 DIVs
+      __m512d e_norm_inv = _mm512_div_pd(_mm512_set1_pd(1.0), e_norm);
+
+      // # 16 MULT
+      __m512d e_x_i_vec = _mm512_mul_pd(ex_vec, e_norm_inv);
+      __m512d e_y_i_vec = _mm512_mul_pd(ey_vec, e_norm_inv);
+
+      // # 16 MULTs, 8 ADDs
+      __m512d vel_norm_sq = _mm512_fmadd_pd(curr_vel_x_i_vec, curr_vel_x_i_vec, _mm512_mul_pd(curr_vel_y_i_vec, curr_vel_y_i_vec));
+
+      // # 8 MULTs
+      __m512d sb_sq_i_vec = _mm512_mul_pd(vel_norm_sq, _mm512_set1_pd(delta_t_sq));
+
+      // # 8 SQRTs
+      __m512d vel_norm_dt = _mm512_sqrt_pd(sb_sq_i_vec);
+
+      _mm512_storeu_pd(&curr_vel_x[i], curr_vel_x_i_vec);
+      _mm512_storeu_pd(&curr_vel_y[i], curr_vel_y_i_vec);
+      _mm512_storeu_pd(&w_x[i], w_x_i_vec);
+      _mm512_storeu_pd(&w_y[i], w_y_i_vec);
+      _mm512_storeu_pd(&pos_x[i], pos_x_i_vec);
+      _mm512_storeu_pd(&pos_y[i], pos_y_i_vec);
+      _mm512_storeu_pd(&e_x[i], e_x_i_vec);
+      _mm512_storeu_pd(&e_y[i], e_y_i_vec);
+      _mm512_storeu_pd(&destination_y[i], destination_y_i_vec);
+      _mm512_storeu_pd(&sb_sq[i], sb_sq_i_vec);
+
+      // # 16 MULTs
+      _mm512_storeu_pd(&sb_e_x[i], _mm512_mul_pd(e_x_i_vec, vel_norm_dt));
+      _mm512_storeu_pd(&sb_e_y[i], _mm512_mul_pd(e_y_i_vec, vel_norm_dt));
+    }
+#if !BENCHMARK
+    for (int i = 0; i < n_p; i++) {
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+    }
+#endif
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp b/variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp
new file mode 100644
index 0000000..486512c
--- /dev/null
+++ b/variants/variant_inlining_8_SoA_AVX512_polyexp_moreavx_ilp_2_expfix.cpp
@@ -0,0 +1,640 @@
+// # Changes: Added ILP with k = 2 to j-loop
+// # Predecessor: variant_inlining_8_SoA_AVX512_polyexp_moreavx.cpp
+
+#include "../headers/board_reduced.hpp"
+#include "../headers/constants.hpp"
+#include "../headers/pedestrian_reduced.hpp"
+
+#include <immintrin.h>
+
+#include <fstream>
+#include <random>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0
+#define BOARD_LENGTH 50.0
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5
+#define WALL_EPS 1e-3
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+alignas(64) vector<double> pos_x, pos_y;
+alignas(64) vector<double> sb_sq;
+alignas(64) vector<double> sb_e_x, sb_e_y;
+alignas(64) vector<double> e_x, e_y;
+alignas(64) vector<double> curr_vel_x, curr_vel_y;
+alignas(64) vector<double> w_x, w_y;
+alignas(64) vector<double> desired_v0;
+alignas(64) vector<double> v_max;
+alignas(64) vector<double> destination_x, destination_y;
+alignas(64) vector<int> color;
+
+Pedestrian *init_pedestrians(int n_p) {
+
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<double> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<double> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<double> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<double> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  const int n_p_pad = (n_p + 15) & ~15;
+
+  pos_x.resize(n_p_pad);
+  pos_y.resize(n_p_pad);
+  sb_sq.resize(n_p_pad);
+  sb_e_x.resize(n_p_pad);
+  sb_e_y.resize(n_p_pad);
+  e_x.resize(n_p_pad);
+  e_y.resize(n_p_pad);
+  curr_vel_x.resize(n_p_pad);
+  curr_vel_y.resize(n_p_pad);
+  w_x.resize(n_p_pad);
+  w_y.resize(n_p_pad);
+  desired_v0.resize(n_p_pad);
+  v_max.resize(n_p_pad);
+  destination_x.resize(n_p_pad);
+  destination_y.resize(n_p_pad);
+  color.resize(n_p_pad);
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    double x = x_gen_left(gen);
+    double y = y_gen(gen);
+
+    double dest_x = BOARD_LENGTH + 30.0;
+    double dest_y = y;
+
+    double init_e_x = dest_x - x;
+    double init_e_y = dest_y - y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1);
+
+    double v_m = 1.3 * d_v0;
+
+    color[i] = 0;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0;
+    w_y[i] = 0.0;
+    curr_vel_x[i] = 0.0;
+    curr_vel_y[i] = 0.0;
+    sb_sq[i] = 0.0;
+    sb_e_x[i] = 0.0;
+    sb_e_y[i] = 0.0;
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    double x = x_gen_right(gen);
+    double y = y_gen(gen);
+
+    double dest_x = -30;
+    double dest_y = y;
+
+    double init_e_x = dest_x - x;
+    double init_e_y = dest_y - y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double d_v0;
+    do {
+      d_v0 = velocity_gen(gen);
+    } while (d_v0 <= 0.1);
+
+    double v_m = 1.3 * d_v0;
+
+    color[i] = 1;
+    pos_x[i] = x;
+    pos_y[i] = y;
+    e_x[i] = init_e_unit_x;
+    e_y[i] = init_e_unit_y;
+    v_max[i] = v_m;
+    desired_v0[i] = d_v0;
+    destination_x[i] = dest_x;
+    destination_y[i] = dest_y;
+
+    w_x[i] = 0.0;
+    w_y[i] = 0.0;
+    curr_vel_x[i] = 0.0;
+    curr_vel_y[i] = 0.0;
+    sb_sq[i] = 0.0;
+    sb_e_x[i] = 0.0;
+    sb_e_y[i] = 0.0;
+  }
+
+  for (int j = n_p; j < n_p_pad; j++) {
+    pos_x[j] = 1e9;
+    pos_y[j] = 1e9;
+    sb_sq[j] = 0.0;
+    sb_e_x[j] = 0.0;
+    sb_e_y[j] = 0.0;
+  }
+  return nullptr;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+// # 128 MULTs, 112 ADDs
+static inline __m512d _avx512_exp_pd(__m512d x) {
+
+  const __m512d log2e = _mm512_set1_pd(1.4426950408889634);
+
+  // # 8 MULTs
+  __m512d nf = _mm512_roundscale_pd(_mm512_mul_pd(x, log2e), _MM_FROUND_TO_NEAREST_INT | _MM_FROUND_NO_EXC);
+  const __m512d ln2 = _mm512_set1_pd(0.6931471805599453);
+
+  // # 8 MULTs, 8 ADDs
+  __m512d r = _mm512_fnmadd_pd(nf, ln2, x);
+
+  const __m512d c13 = _mm512_set1_pd(c13_scalar);
+  const __m512d c12 = _mm512_set1_pd(c12_scalar);
+  const __m512d c11 = _mm512_set1_pd(c11_scalar);
+  const __m512d c10 = _mm512_set1_pd(c10_scalar);
+  const __m512d c9 = _mm512_set1_pd(c9_scalar);
+  const __m512d c8 = _mm512_set1_pd(c8_scalar);
+  const __m512d c7 = _mm512_set1_pd(c7_scalar);
+  const __m512d c6 = _mm512_set1_pd(c6_scalar);
+  const __m512d c5 = _mm512_set1_pd(c5_scalar);
+  const __m512d c4 = _mm512_set1_pd(c4_scalar);
+  const __m512d c3 = _mm512_set1_pd(c3_scalar);
+  const __m512d c2 = _mm512_set1_pd(c2_scalar);
+  const __m512d c1 = _mm512_set1_pd(c1_scalar);
+  const __m512d c0 = _mm512_set1_pd(c0_scalar);
+
+  __m512d p = c13;
+
+  // # 104 MULTs, 104 ADDs
+  p = _mm512_fmadd_pd(p, r, c12);
+  p = _mm512_fmadd_pd(p, r, c11);
+  p = _mm512_fmadd_pd(p, r, c10);
+  p = _mm512_fmadd_pd(p, r, c9);
+  p = _mm512_fmadd_pd(p, r, c8);
+  p = _mm512_fmadd_pd(p, r, c7);
+  p = _mm512_fmadd_pd(p, r, c6);
+  p = _mm512_fmadd_pd(p, r, c5);
+  p = _mm512_fmadd_pd(p, r, c4);
+  p = _mm512_fmadd_pd(p, r, c3);
+  p = _mm512_fmadd_pd(p, r, c2);
+  p = _mm512_fmadd_pd(p, r, c1);
+  p = _mm512_fmadd_pd(p, r, c0);
+
+  __m256d nf_lo = _mm512_castpd512_pd256(nf);
+  __m256d nf_hi = _mm512_extractf64x4_pd(nf, 1);
+
+  __m128i ni32_lo = _mm256_cvtpd_epi32(nf_lo);
+  __m128i ni32_hi = _mm256_cvtpd_epi32(nf_hi);
+
+  __m256i ni64_lo = _mm256_cvtepi32_epi64(ni32_lo);
+  __m256i ni64_hi = _mm256_cvtepi32_epi64(ni32_hi);
+
+  __m512i ni64 = _mm512_inserti64x4(_mm512_castsi256_si512(ni64_lo), ni64_hi, 1);
+
+  __m512i biased = _mm512_add_epi64(ni64, _mm512_set1_epi64(1023));
+
+  __m512i scaled = _mm512_slli_epi64(biased, 52);
+
+  __m512d pow2n = _mm512_castsi512_pd(scaled);
+
+  // # 8 MULTs
+  return _mm512_mul_pd(p, pow2n);
+}
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+
+  vector<tuple<int, int, double, double, double, double, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+
+    for (int i = 0; i < n_p; i++) {
+      samples.emplace_back(0, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+    }
+  }
+#endif
+
+  const int n_p_pad = (n_p + 15) & ~15;
+
+  alignas(64) constexpr double lane_offset_arr[8] = {0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0};
+  const __m512d lane_off = _mm512_load_pd(lane_offset_arr);
+
+  const __m512d v_neg_v0_div_2s = _mm512_set1_pd(NEG_V0_AB_div_TWO_SIGMA);
+  const __m512d v_neg_2s_inv = _mm512_set1_pd(NEG_TWO_SIGMA_INV);
+  const __m512d v_cos_phi_sq = _mm512_set1_pd(cos_phi_sq);
+  const __m512d v_dt_neg = _mm512_set1_pd(dt_neg);
+  const __m512d v_c_neg_dt = _mm512_set1_pd(c_neg_times_dt);
+  const __m512d v_zero = _mm512_setzero_pd();
+  const __m512d v_one = _mm512_set1_pd(1.0f);
+  const __m512d v_np = _mm512_set1_pd((float)n_p);
+  const __m512d v_exp_min = _mm512_set1_pd(-87.0f);
+
+  // # ADDs: n_iter * n_p * 36 +
+  // #        n_iter * n_p * (n_p_pad / 16) * 464 +
+  // #        n_iter * n_p * board.n_obstacles * 10 +
+  // #        n_iter * (n_p_pad / 8) * 56
+
+  // # MULTs: n_iter * n_p * 4 +
+  // #        n_iter * n_p * (n_p_pad / 16) * 640 +
+  // #        n_iter * n_p * board.n_obstacles * 13 +
+  // #        n_iter * (n_p_pad / 8) * 120
+
+  // # DIVs: n_iter * n_p * (n_p_pad / 16) * 16 +
+  // #        n_iter * n_p * board.n_obstacles +
+  // #        n_iter * (n_p_pad / 8) * 16
+
+  // # SQRTs: n_iter * n_p * (n_p_pad / 16) * 48 +
+  // #        n_iter * n_p * board.n_obstacles +
+  // #        n_iter * (n_p_pad / 8) * 24
+
+  // # EXPs: n_iter * n_p * board.n_obstacles
+
+  for (int t = 0; t < n_iter; t++) {
+
+    for (int i = 0; i < n_p; i++) {
+
+      double pos_x_i = pos_x[i];
+      double pos_y_i = pos_y[i];
+      double e_x_i = e_x[i];
+      double e_y_i = e_y[i];
+      double desired_v0_i = desired_v0[i];
+      double curr_vel_x_i = curr_vel_x[i];
+      double curr_vel_y_i = curr_vel_y[i];
+
+      // # 4 MULTs, 2 ADDs
+      double f_alpha_dt_x = (e_x_i * desired_v0_i - curr_vel_x_i) * dt_times_tau_alpha_inv;
+      double f_alpha_dt_y = (e_y_i * desired_v0_i - curr_vel_y_i) * dt_times_tau_alpha_inv;
+
+      const __m512d pix = _mm512_set1_pd(pos_x_i);
+      const __m512d piy = _mm512_set1_pd(pos_y_i);
+      const __m512d eix = _mm512_set1_pd(e_x_i);
+      const __m512d eiy = _mm512_set1_pd(e_y_i);
+      const __m512d v_i = _mm512_set1_pd((double)i);
+
+      __m512d acc_x_0 = _mm512_setzero_pd();
+      __m512d acc_y_0 = _mm512_setzero_pd();
+      __m512d acc_x_1 = _mm512_setzero_pd();
+      __m512d acc_y_1 = _mm512_setzero_pd();
+
+      for (int j = 0; j < n_p_pad; j += 16) {
+
+        // # 16 ADDs
+        __m512d j_idx_0 = _mm512_add_pd(_mm512_set1_pd((double)j), lane_off);
+        __m512d j_idx_1 = _mm512_add_pd(_mm512_set1_pd((double)j + 8), lane_off);
+
+        __mmask8 m_neq_i_0 = _mm512_cmp_pd_mask(j_idx_0, v_i, _CMP_NEQ_OQ);
+        __mmask8 m_neq_i_1 = _mm512_cmp_pd_mask(j_idx_1, v_i, _CMP_NEQ_OQ);
+
+        __mmask8 m_in_range_0 = _mm512_cmp_pd_mask(j_idx_0, v_np, _CMP_LT_OQ);
+        __mmask8 m_in_range_1 = _mm512_cmp_pd_mask(j_idx_1, v_np, _CMP_LT_OQ);
+
+        __mmask8 active_0 = m_neq_i_0 & m_in_range_0;
+        __mmask8 active_1 = m_neq_i_1 & m_in_range_1;
+
+        __m512d pjx_0 = _mm512_loadu_pd(&pos_x[j]);
+        __m512d pjx_1 = _mm512_loadu_pd(&pos_x[j + 8]);
+
+        __m512d pjy_0 = _mm512_loadu_pd(&pos_y[j]);
+        __m512d pjy_1 = _mm512_loadu_pd(&pos_y[j + 8]);
+
+        __m512d sbe_jx_0 = _mm512_loadu_pd(&sb_e_x[j]);
+        __m512d sbe_jx_1 = _mm512_loadu_pd(&sb_e_x[j + 8]);
+
+        __m512d sbe_jy_0 = _mm512_loadu_pd(&sb_e_y[j]);
+        __m512d sbe_jy_1 = _mm512_loadu_pd(&sb_e_y[j + 8]);
+
+        __m512d sb_jsq_0 = _mm512_loadu_pd(&sb_sq[j]);
+        __m512d sb_jsq_1 = _mm512_loadu_pd(&sb_sq[j + 8]);
+
+        // # 16 ADDs
+        __m512d r_ab_x_0 = _mm512_sub_pd(pix, pjx_0);
+        __m512d r_ab_x_1 = _mm512_sub_pd(pix, pjx_1);
+
+        // # 16 ADDs
+        __m512d r_ab_y_0 = _mm512_sub_pd(piy, pjy_0);
+        __m512d r_ab_y_1 = _mm512_sub_pd(piy, pjy_1);
+
+        // # 32 MULTs, 16 ADDs
+        __m512d r_ab_n2_0 = _mm512_fmadd_pd(r_ab_x_0, r_ab_x_0, _mm512_mul_pd(r_ab_y_0, r_ab_y_0));
+        __m512d r_ab_n2_1 = _mm512_fmadd_pd(r_ab_x_1, r_ab_x_1, _mm512_mul_pd(r_ab_y_1, r_ab_y_1));
+
+        // # 16 SQRTs
+        __m512d r_ab_n_0 = _mm512_sqrt_pd(r_ab_n2_0);
+        __m512d r_ab_n_1 = _mm512_sqrt_pd(r_ab_n2_1);
+
+        // # 16 ADDs
+        __m512d r_ab2_x_0 = _mm512_sub_pd(r_ab_x_0, sbe_jx_0);
+        __m512d r_ab2_x_1 = _mm512_sub_pd(r_ab_x_1, sbe_jx_1);
+
+        // # 16 ADDs
+        __m512d r_ab2_y_0 = _mm512_sub_pd(r_ab_y_0, sbe_jy_0);
+        __m512d r_ab2_y_1 = _mm512_sub_pd(r_ab_y_1, sbe_jy_1);
+
+        // # 32 MULTs, 16 ADDs
+        __m512d r_ab2_n2_0 = _mm512_fmadd_pd(r_ab2_x_0, r_ab2_x_0, _mm512_mul_pd(r_ab2_y_0, r_ab2_y_0));
+        __m512d r_ab2_n2_1 = _mm512_fmadd_pd(r_ab2_x_1, r_ab2_x_1, _mm512_mul_pd(r_ab2_y_1, r_ab2_y_1));
+
+        // # 16 SQRTs
+        __m512d r_ab2_n_0 = _mm512_sqrt_pd(r_ab2_n2_0);
+        __m512d r_ab2_n_1 = _mm512_sqrt_pd(r_ab2_n2_1);
+
+        // # 16 ADDs
+        __m512d r_sum_0 = _mm512_add_pd(r_ab_n_0, r_ab2_n_0);
+        __m512d r_sum_1 = _mm512_add_pd(r_ab_n_1, r_ab2_n_1);
+
+        // # 16 MULTs, 16 ADDs
+        __m512d two_b_sq_0 = _mm512_fmsub_pd(r_sum_0, r_sum_0, sb_jsq_0);
+        __m512d two_b_sq_1 = _mm512_fmsub_pd(r_sum_1, r_sum_1, sb_jsq_1);
+
+        // # 16 SQRTs
+        __m512d two_b_0 = _mm512_sqrt_pd(two_b_sq_0);
+        __m512d two_b_1 = _mm512_sqrt_pd(two_b_sq_1);
+
+        // # 16 MULTs
+        __m512d e_arg_0 = _mm512_mul_pd(two_b_0, v_neg_2s_inv);
+        __m512d e_arg_1 = _mm512_mul_pd(two_b_1, v_neg_2s_inv);
+
+        e_arg_0 = _mm512_max_pd(e_arg_0, v_exp_min);
+        e_arg_1 = _mm512_max_pd(e_arg_1, v_exp_min);
+
+        // # 256 MULTs, 224 ADDs
+        __m512d e_val_0 = _avx512_exp_pd(e_arg_0);
+        __m512d e_val_1 = _avx512_exp_pd(e_arg_1);
+
+        // # 32 MULTs
+        __m512d numer_0 = _mm512_mul_pd(_mm512_mul_pd(r_sum_0, v_neg_v0_div_2s), e_val_0);
+        __m512d numer_1 = _mm512_mul_pd(_mm512_mul_pd(r_sum_1, v_neg_v0_div_2s), e_val_1);
+
+        // # 32 MULTs
+        __m512d denom_0 = _mm512_mul_pd(_mm512_mul_pd(two_b_0, r_ab_n_0), r_ab2_n_0);
+        __m512d denom_1 = _mm512_mul_pd(_mm512_mul_pd(two_b_1, r_ab_n_1), r_ab2_n_1);
+
+        __m512d denom_safe_0 = _mm512_mask_blend_pd(active_0, v_one, denom_0);
+        __m512d denom_safe_1 = _mm512_mask_blend_pd(active_1, v_one, denom_1);
+
+        // # 16 DIVs
+        __m512d smc_0 = _mm512_div_pd(numer_0, denom_safe_0);
+        __m512d smc_1 = _mm512_div_pd(numer_1, denom_safe_1);
+
+        // # 16 MULTs
+        __m512d sm_r_ab_0 = _mm512_mul_pd(smc_0, r_ab2_n_0);
+        __m512d sm_r_ab_1 = _mm512_mul_pd(smc_1, r_ab2_n_1);
+
+        // # 16 MULTs
+        __m512d sm_r_ab2_0 = _mm512_mul_pd(smc_0, r_ab_n_0);
+        __m512d sm_r_ab2_1 = _mm512_mul_pd(smc_1, r_ab_n_1);
+
+        // # 32 MULTs, 16 ADDs
+        __m512d f_x_0 = _mm512_fmadd_pd(r_ab2_x_0, sm_r_ab2_0, _mm512_mul_pd(r_ab_x_0, sm_r_ab_0));
+        __m512d f_x_1 = _mm512_fmadd_pd(r_ab2_x_1, sm_r_ab2_1, _mm512_mul_pd(r_ab_x_1, sm_r_ab_1));
+
+        // # 32 MULTs, 16 ADDs
+        __m512d f_y_0 = _mm512_fmadd_pd(r_ab2_y_0, sm_r_ab2_0, _mm512_mul_pd(r_ab_y_0, sm_r_ab_0));
+        __m512d f_y_1 = _mm512_fmadd_pd(r_ab2_y_1, sm_r_ab2_1, _mm512_mul_pd(r_ab_y_1, sm_r_ab_1));
+
+        // # 32 MULTs, 16 ADDs
+        __m512d f_norm_sq_0 = _mm512_fmadd_pd(f_x_0, f_x_0, _mm512_mul_pd(f_y_0, f_y_0));
+        __m512d f_norm_sq_1 = _mm512_fmadd_pd(f_x_1, f_x_1, _mm512_mul_pd(f_y_1, f_y_1));
+
+        // # 16 MULTs
+        __m512d rhs_0 = _mm512_mul_pd(f_norm_sq_0, v_cos_phi_sq);
+        __m512d rhs_1 = _mm512_mul_pd(f_norm_sq_1, v_cos_phi_sq);
+
+        // # 32 MULTs, 16 ADDs
+        __m512d e_dot_f_0 = _mm512_fmadd_pd(eix, f_x_0, _mm512_mul_pd(eiy, f_y_0));
+        __m512d e_dot_f_1 = _mm512_fmadd_pd(eix, f_x_1, _mm512_mul_pd(eiy, f_y_1));
+
+        // # 16 MULTs
+        __m512d e_dot_f_sq_0 = _mm512_mul_pd(e_dot_f_0, e_dot_f_0);
+        __m512d e_dot_f_sq_1 = _mm512_mul_pd(e_dot_f_1, e_dot_f_1);
+
+        __mmask8 c1_0 = _mm512_cmp_pd_mask(e_dot_f_0, v_zero, _CMP_LT_OQ);
+        __mmask8 c1_1 = _mm512_cmp_pd_mask(e_dot_f_1, v_zero, _CMP_LT_OQ);
+
+        __mmask8 c2_0 = _mm512_cmp_pd_mask(e_dot_f_sq_0, rhs_0, _CMP_GT_OQ);
+        __mmask8 c2_1 = _mm512_cmp_pd_mask(e_dot_f_sq_1, rhs_1, _CMP_GT_OQ);
+
+        __mmask8 cond_0 = c1_0 & c2_0;
+        __mmask8 cond_1 = c1_1 & c2_1;
+
+        __m512d w_val_0 = _mm512_mask_blend_pd(cond_0, v_dt_neg, v_c_neg_dt);
+        __m512d w_val_1 = _mm512_mask_blend_pd(cond_1, v_dt_neg, v_c_neg_dt);
+
+        __m512d f_x_safe_0 = _mm512_maskz_mov_pd(active_0, f_x_0);
+        __m512d f_x_safe_1 = _mm512_maskz_mov_pd(active_1, f_x_1);
+
+        __m512d f_y_safe_0 = _mm512_maskz_mov_pd(active_0, f_y_0);
+        __m512d f_y_safe_1 = _mm512_maskz_mov_pd(active_1, f_y_1);
+
+        // # 16 MULTs, 16 ADDs
+        acc_x_0 = _mm512_fmadd_pd(f_x_safe_0, w_val_0, acc_x_0);
+        acc_x_1 = _mm512_fmadd_pd(f_x_safe_1, w_val_1, acc_x_1);
+
+        // # 16 MULTs, 16 ADDs
+        acc_y_0 = _mm512_fmadd_pd(f_y_safe_0, w_val_0, acc_y_0);
+        acc_y_1 = _mm512_fmadd_pd(f_y_safe_1, w_val_1, acc_y_1);
+      }
+
+      // # 32 ADDs
+      f_alpha_dt_x += _mm512_reduce_add_pd(_mm512_add_pd(acc_x_0, acc_x_1));
+      f_alpha_dt_y += _mm512_reduce_add_pd(_mm512_add_pd(acc_y_0, acc_y_1));
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
+        double v1_x = pos_x_i - board.obstacles[k].start_x;
+        double v1_y = pos_y_i - board.obstacles[k].start_y;
+
+        // # 3 MULTs, 1 ADD
+        double projection = (v1_x * board.obstacles[k].obstacle_v_x + v1_y * board.obstacles[k].obstacle_v_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0)
+          projection = 0.0;
+        else if (projection > 1.0)
+          projection = 1.0;
+
+        // # 4 ADDs, 2 MULTs
+        double r_aB_x = pos_x_i - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        double r_aB_y = pos_y_i - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        double r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+
+        // # 2 MULTs, 1 EXP, 1 DIV
+        double scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+
+        // # 2 MULTs
+        double f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        double f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        double w_val = dt_neg;
+        double e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -e_y_i : e_y_i);
+        if (e_dot_f < cos_phi)
+          w_val = c_neg_times_dt;
+
+        // # 2 MULTs, 2 ADDs
+        f_alpha_dt_x += f_alpha_B_neg_x * w_val;
+        f_alpha_dt_y += f_alpha_B_neg_y * w_val;
+      }
+
+      // # 2 ADDs
+      w_x[i] += f_alpha_dt_x;
+      w_y[i] += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p_pad; i += 8) {
+
+      __m512d w_x_i_vec = _mm512_loadu_pd(&w_x[i]);
+      __m512d w_y_i_vec = _mm512_loadu_pd(&w_y[i]);
+      __m512d v_max_i_vec = _mm512_loadu_pd(&v_max[i]);
+      __m512d pos_x_i_vec = _mm512_loadu_pd(&pos_x[i]);
+      __m512d pos_y_i_vec = _mm512_loadu_pd(&pos_y[i]);
+      __m512d destination_y_i_vec = _mm512_loadu_pd(&destination_y[i]);
+      __m512d destination_x_i_vec = _mm512_loadu_pd(&destination_x[i]);
+      __m512d desired_v0_i_vec = _mm512_loadu_pd(&desired_v0[i]);
+
+      __m256i color_i_vec = _mm256_loadu_si256((__m256i *)&color[i]);
+
+      // # 16 MULTs, 8 ADDs
+      __m512d w_norm_sq = _mm512_fmadd_pd(w_x_i_vec, w_x_i_vec, _mm512_mul_pd(w_y_i_vec, w_y_i_vec));
+
+      // # 8 SQRTs
+      __m512d w_norm = _mm512_sqrt_pd(w_norm_sq);
+
+      // # 8 DIVs
+      __m512d correction_factor = _mm512_min_pd(_mm512_set1_pd(1.0), _mm512_div_pd(v_max_i_vec, w_norm));
+
+      // # 16 MULTs
+      __m512d curr_vel_x_i_vec = _mm512_mul_pd(w_x_i_vec, correction_factor);
+      __m512d curr_vel_y_i_vec = _mm512_mul_pd(w_y_i_vec, correction_factor);
+
+      __m512d dt_vec = _mm512_set1_pd(dt);
+
+      // # 16 MULTs, 16 ADDs
+      pos_x_i_vec = _mm512_fmadd_pd(curr_vel_x_i_vec, dt_vec, pos_x_i_vec);
+      pos_y_i_vec = _mm512_fmadd_pd(curr_vel_y_i_vec, dt_vec, pos_y_i_vec);
+
+      __mmask8 cond_1_left = _mm256_cmpeq_epi32_mask(color_i_vec, _mm256_setzero_si256());
+      __mmask8 cond_1_right = _mm512_cmp_pd_mask(pos_x_i_vec, _mm512_set1_pd(BOARD_LENGTH_MINUS_ONE_POINT_FIVE), _CMP_GT_OQ);
+      __mmask8 cond_1 = cond_1_left & cond_1_right;
+
+      pos_x_i_vec = _mm512_mask_blend_pd(cond_1, pos_x_i_vec, _mm512_set1_pd(1.0));
+      destination_y_i_vec = _mm512_mask_blend_pd(cond_1, destination_y_i_vec, pos_y_i_vec);
+      w_x_i_vec = _mm512_mask_blend_pd(cond_1, w_x_i_vec, desired_v0_i_vec);
+      w_y_i_vec = _mm512_mask_blend_pd(cond_1, w_y_i_vec, _mm512_setzero_pd());
+      curr_vel_x_i_vec = _mm512_mask_blend_pd(cond_1, curr_vel_x_i_vec, w_x_i_vec);
+      curr_vel_y_i_vec = _mm512_mask_blend_pd(cond_1, curr_vel_y_i_vec, w_y_i_vec);
+
+      __mmask8 cond_2_left = _mm256_cmpeq_epi32_mask(color_i_vec, _mm256_set1_epi32(1));
+      __mmask8 cond_2_right = _mm512_cmp_pd_mask(pos_x_i_vec, _mm512_set1_pd(1.5), _CMP_LT_OQ);
+      __mmask8 cond_2 = cond_2_left & cond_2_right;
+
+      pos_x_i_vec = _mm512_mask_blend_pd(cond_2, pos_x_i_vec, _mm512_set1_pd(BOARD_LENGTH - 1.0));
+      destination_y_i_vec = _mm512_mask_blend_pd(cond_2, destination_y_i_vec, pos_y_i_vec);
+      __m512d neg_desired_v0 = _mm512_xor_pd(desired_v0_i_vec, _mm512_set1_pd(-0.0));
+      w_x_i_vec = _mm512_mask_blend_pd(cond_2, w_x_i_vec, neg_desired_v0);
+      w_y_i_vec = _mm512_mask_blend_pd(cond_2, w_y_i_vec, _mm512_setzero_pd());
+      curr_vel_x_i_vec = _mm512_mask_blend_pd(cond_2, curr_vel_x_i_vec, w_x_i_vec);
+      curr_vel_y_i_vec = _mm512_mask_blend_pd(cond_2, curr_vel_y_i_vec, w_y_i_vec);
+
+      pos_y_i_vec = _mm512_max_pd(pos_y_i_vec, _mm512_set1_pd(WALL_EPS));
+      pos_y_i_vec = _mm512_min_pd(pos_y_i_vec, _mm512_set1_pd(BOARD_WIDTH_MINUS_WALL_EPS));
+
+      // # 16 ADDs
+      __m512d ex_vec = _mm512_sub_pd(destination_x_i_vec, pos_x_i_vec);
+      __m512d ey_vec = _mm512_sub_pd(destination_y_i_vec, pos_y_i_vec);
+
+      // # 16 MULTs, 8 ADDs
+      __m512d e_norm_sq = _mm512_fmadd_pd(ex_vec, ex_vec, _mm512_mul_pd(ey_vec, ey_vec));
+
+      // # 8 SQRTs
+      __m512d e_norm = _mm512_sqrt_pd(e_norm_sq);
+
+      // # 8 DIVs
+      __m512d e_norm_inv = _mm512_div_pd(_mm512_set1_pd(1.0), e_norm);
+
+      // # 16 MULT
+      __m512d e_x_i_vec = _mm512_mul_pd(ex_vec, e_norm_inv);
+      __m512d e_y_i_vec = _mm512_mul_pd(ey_vec, e_norm_inv);
+
+      // # 16 MULTs, 8 ADDs
+      __m512d vel_norm_sq = _mm512_fmadd_pd(curr_vel_x_i_vec, curr_vel_x_i_vec, _mm512_mul_pd(curr_vel_y_i_vec, curr_vel_y_i_vec));
+
+      // # 8 MULTs
+      __m512d sb_sq_i_vec = _mm512_mul_pd(vel_norm_sq, _mm512_set1_pd(delta_t_sq));
+
+      // # 8 SQRTs
+      __m512d vel_norm_dt = _mm512_sqrt_pd(sb_sq_i_vec);
+
+      _mm512_storeu_pd(&curr_vel_x[i], curr_vel_x_i_vec);
+      _mm512_storeu_pd(&curr_vel_y[i], curr_vel_y_i_vec);
+      _mm512_storeu_pd(&w_x[i], w_x_i_vec);
+      _mm512_storeu_pd(&w_y[i], w_y_i_vec);
+      _mm512_storeu_pd(&pos_x[i], pos_x_i_vec);
+      _mm512_storeu_pd(&pos_y[i], pos_y_i_vec);
+      _mm512_storeu_pd(&e_x[i], e_x_i_vec);
+      _mm512_storeu_pd(&e_y[i], e_y_i_vec);
+      _mm512_storeu_pd(&destination_y[i], destination_y_i_vec);
+      _mm512_storeu_pd(&sb_sq[i], sb_sq_i_vec);
+
+      // # 16 MULTs
+      _mm512_storeu_pd(&sb_e_x[i], _mm512_mul_pd(e_x_i_vec, vel_norm_dt));
+      _mm512_storeu_pd(&sb_e_y[i], _mm512_mul_pd(e_y_i_vec, vel_norm_dt));
+    }
+#if !BENCHMARK
+    for (int i = 0; i < n_p; i++) {
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, pos_x[i], pos_y[i], e_x[i], e_y[i], color[i]);
+      }
+    }
+#endif
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}
diff --git a/variants/variant_inlining_8_flopcount.cpp b/variants/variant_inlining_8_flopcount.cpp
new file mode 100644
index 0000000..45ffc24
--- /dev/null
+++ b/variants/variant_inlining_8_flopcount.cpp
@@ -0,0 +1,343 @@
+// # This file performs in-lining for run_simulation.
+// # Changes: Replaced references to Vector2d structs inside each Obstacle of Board.obstacles with references to doubles
+// # Predecessor: variant_inlining_7.cpp
+
+#include "../headers/board_reduced.hpp"
+#include "../headers/constants.hpp"
+#include "../headers/pedestrian_reduced.hpp"
+
+#include <algorithm>
+#include <cmath>
+#include <fstream>
+#include <iostream>
+#include <math.h>
+#include <random>
+#include <tuple>
+#include <vector>
+
+#define SEED 42
+#define BOARD_WIDTH 10.0
+#define BOARD_LENGTH 50.0
+#define BOARD_LENGTH_MINUS_ONE BOARD_LENGTH - 1.0
+#define BOARD_LENGTH_MINUS_ONE_POINT_FIVE BOARD_LENGTH - 1.5
+#define WALL_EPS 1e-3
+#define BOARD_WIDTH_MINUS_WALL_EPS 9.999
+
+#ifndef RANDOM_SEED
+#define RANDOM_SEED 0
+#endif
+
+#ifndef BENCHMARK
+#define BENCHMARK 0
+#endif
+
+using namespace std;
+
+Pedestrian *init_pedestrians(int n_p) {
+#if RANDOM_SEED
+  random_device rd;
+  mt19937 gen(rd());
+#else
+  mt19937 gen(SEED);
+#endif
+
+  normal_distribution<double> velocity_gen(v0_mean, v0_std_dev);
+  uniform_real_distribution<double> y_gen(0.3, BOARD_WIDTH - 0.3);
+  uniform_real_distribution<double> x_gen_left(2.0, BOARD_LENGTH / 2.0);
+  uniform_real_distribution<double> x_gen_right(BOARD_LENGTH / 2.0, BOARD_LENGTH - 2.0);
+
+  Pedestrian *P = new Pedestrian[n_p];
+
+  int n_p_half = n_p / 2;
+
+  for (int i = 0; i < n_p_half; i++) {
+    double x = x_gen_left(gen);
+    double y = y_gen(gen);
+
+    double start_pos_x = x;
+    double start_pos_y = y;
+
+    double destination_x = BOARD_LENGTH + 30.0;
+    double destination_y = y;
+
+    double init_e_x = destination_x - start_pos_x;
+    double init_e_y = destination_y - start_pos_y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double desired_v0;
+    do {
+      desired_v0 = velocity_gen(gen);
+    } while (desired_v0 <= 0.1);
+
+    double v_max = 1.3 * desired_v0;
+
+    P[i] = Pedestrian(0, 0, 0, start_pos_x, start_pos_y, init_e_unit_x, init_e_unit_y, v_max, desired_v0, destination_x, destination_y);
+  }
+
+  for (int i = n_p_half; i < n_p; i++) {
+    double x = x_gen_right(gen);
+    double y = y_gen(gen);
+
+    double start_pos_x = x;
+    double start_pos_y = y;
+    double destination_x = -30;
+    double destination_y = y;
+
+    double init_e_x = destination_x - start_pos_x;
+    double init_e_y = destination_y - start_pos_y;
+    double init_e_norm = sqrt(init_e_x * init_e_x + init_e_y * init_e_y);
+    double init_e_unit_x = init_e_x / init_e_norm;
+    double init_e_unit_y = init_e_y / init_e_norm;
+
+    double desired_v0;
+    do {
+      desired_v0 = velocity_gen(gen);
+    } while (desired_v0 <= 0.1);
+
+    double v_max = 1.3 * desired_v0;
+
+    P[i] = Pedestrian(1, 0, 0, start_pos_x, start_pos_y, init_e_unit_x, init_e_unit_y, v_max, desired_v0, destination_x, destination_y);
+  }
+
+  return P;
+}
+
+Board init_board_without_obstacles() { return Board(BOARD_LENGTH, BOARD_WIDTH, 0, nullptr); }
+
+int run_simulation(Pedestrian *pedestrians, Board &board, int n_p, int n_iter, const char *dump_path) {
+
+#if BENCHMARK
+  (void)dump_path;
+#endif
+
+#if !BENCHMARK
+  const int dump_interval = 1;
+
+  vector<tuple<int, int, double, double, double, double, int>> samples;
+
+  if (dump_path) {
+    samples.reserve(static_cast<size_t>(n_iter / dump_interval + 1) * n_p);
+
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+      samples.emplace_back(0, i, p.curr_pos_x, p.curr_pos_y, p.e_x, p.e_y, p.color);
+    }
+  }
+#endif
+
+  // # ADDs: n_iter * n_p * 11 +
+  // #        n_iter * n_p * (n_p - 1) * 14 +
+  // #        n_iter * n_p * board.n_obstacles * 10
+
+  // # MULTs: n_iter * n_p * 21 +
+  // #        n_iter * n_p * (n_p - 1) * 24 +
+  // #        n_iter * n_p * board.n_obstacles * 13
+
+  // # DIVs: n_iter * n_p * 2 +
+  // #        n_iter * n_p * (n_p - 1) +
+  // #        n_iter * n_p * board.n_obstacles
+
+  // # SQRTs: n_iter * n_p * 3 +
+  // #        n_iter * n_p * (n_p - 1) * 3 +
+  // #        n_iter * n_p * board.n_obstacles
+
+  // # EXPs: n_iter * n_p * (n_p - 1) +
+  // #       n_iter * n_p * board.n_obstacles
+
+  for (int t = 0; t < n_iter; t++) {
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+
+      // # 2 MULTs
+      double desired_velocity_x = p.e_x * p.desired_v0;
+      double desired_velocity_y = p.e_y * p.desired_v0;
+
+      // # 2 MULTs, 2 ADDs
+      double f_alpha_dt_x = (desired_velocity_x - p.curr_vel_x) * dt_times_tau_alpha_inv;
+      double f_alpha_dt_y = (desired_velocity_y - p.curr_vel_y) * dt_times_tau_alpha_inv;
+
+      for (int j = 0; j < n_p; j++) {
+        if (i == j)
+          continue;
+
+        // # 2 ADDs
+        double r_ab_x = p.curr_pos_x - pedestrians[j].curr_pos_x;
+        double r_ab_y = p.curr_pos_y - pedestrians[j].curr_pos_y;
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        double r_ab_norm = sqrt(r_ab_x * r_ab_x + r_ab_y * r_ab_y);
+
+        // # 2 ADDs
+        double r_ab2_x = r_ab_x - pedestrians[j].s_b_e_x;
+        double r_ab2_y = r_ab_y - pedestrians[j].s_b_e_y;
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        double r_ab2_norm = sqrt(r_ab2_x * r_ab2_x + r_ab2_y * r_ab2_y);
+
+        // # 1ADD
+        double r_ab_norm_sum = r_ab_norm + r_ab2_norm;
+
+        // # 1 MULT, 1 ADD
+        double two_b_sq = r_ab_norm_sum * r_ab_norm_sum - pedestrians[j].s_b_sq;
+
+        // # 1 SQRT
+        double two_b = sqrt(two_b_sq);
+
+        // # 5 MULTs, 1 EXP, 1 DIV
+        double scalar_mult_common = (r_ab_norm_sum / (two_b * r_ab_norm * r_ab2_norm)) * (NEG_V0_AB_div_TWO_SIGMA * exp(two_b * NEG_TWO_SIGMA_INV));
+
+        // # 2 MULTs
+        double scalar_mult_r_ab = scalar_mult_common * r_ab2_norm;
+        double scalar_mult_r_ab2 = scalar_mult_common * r_ab_norm;
+
+        // # 4 MULTs, 2 ADDs
+        double f_alpha_beta_neg_x = r_ab_x * scalar_mult_r_ab + r_ab2_x * scalar_mult_r_ab2;
+        double f_alpha_beta_neg_y = r_ab_y * scalar_mult_r_ab + r_ab2_y * scalar_mult_r_ab2;
+
+        double w_val;
+
+        // # 3 MULTs, 1 ADD
+        double a = (p.e_x * f_alpha_beta_neg_x + p.e_y * f_alpha_beta_neg_y) * cos_phi_inv;
+
+        // # 2 MULTs, 1 ADD
+        double b_sq = f_alpha_beta_neg_x * f_alpha_beta_neg_x + f_alpha_beta_neg_y * f_alpha_beta_neg_y;
+
+        w_val = dt_neg;
+
+        // # 1 MULT
+        if (a >= 0 && (a * a > b_sq)) {
+          w_val = c_neg_times_dt;
+        }
+
+        // # 2 MULTs, 2 ADDs
+        f_alpha_dt_x = f_alpha_dt_x + f_alpha_beta_neg_x * w_val;
+        f_alpha_dt_y = f_alpha_dt_y + f_alpha_beta_neg_y * w_val;
+      }
+
+      for (int k = 0; k < board.n_obstacles; k++) {
+
+        // # 2 ADDs
+        double v1_x = p.curr_pos_x - board.obstacles[k].start_x;
+        double v1_y = p.curr_pos_y - board.obstacles[k].start_y;
+
+        // # 3 MULTs, 1 ADD
+        double projection = (v1_x * board.obstacles[k].obstacle_v_x + v1_y * board.obstacles[k].obstacle_v_y) * board.obstacles[k].length_sq_inv;
+
+        if (projection < 0.0)
+          projection = 0.0;
+        else if (projection > 1.0)
+          projection = 1.0;
+
+        // # 4 ADDs, 2 MULTs
+        double r_aB_x = p.curr_pos_x - (board.obstacles[k].start_x + board.obstacles[k].obstacle_v_x * projection);
+        double r_aB_y = p.curr_pos_y - (board.obstacles[k].start_y + board.obstacles[k].obstacle_v_y * projection);
+
+        // # 1 SQRT, 2 MULTs, 1 ADD
+        double r_aB_norm = sqrt(r_aB_x * r_aB_x + r_aB_y * r_aB_y);
+        // # 4 MULTs, 1 EXP, 1 DIV
+        double scalar_mult = (NEG_U0_AB_div_R * exp(r_aB_norm * NEG_R_INV)) / r_aB_norm;
+        double f_alpha_B_neg_x = r_aB_x * scalar_mult;
+        double f_alpha_B_neg_y = r_aB_y * scalar_mult;
+
+        double w_val;
+        double e_dot_f = (f_alpha_B_neg_y == 0) ? 0 : (f_alpha_B_neg_y < 0 ? -p.e_y : p.e_y);
+        if (e_dot_f >= cos_phi)
+          w_val = dt_neg;
+        else
+          w_val = c_neg_times_dt;
+
+        // # 2 MULTs
+        double f_alpha_B_dt_x = f_alpha_B_neg_x * w_val;
+        double f_alpha_B_dt_y = f_alpha_B_neg_y * w_val;
+
+        // # 2 ADDs
+        f_alpha_dt_x += f_alpha_B_dt_x;
+        f_alpha_dt_y += f_alpha_B_dt_y;
+      }
+
+      // # 2 ADDs
+      p.w_x += f_alpha_dt_x;
+      p.w_y += f_alpha_dt_y;
+    }
+
+    for (int i = 0; i < n_p; i++) {
+      Pedestrian &p = pedestrians[i];
+
+      // # 1 SQRT, 2 MULTs, 1 ADD
+      double w_norm = sqrt(p.w_x * p.w_x + p.w_y * p.w_y);
+
+      // # 1 DIV
+      double correction_factor = min(1.0, p.v_max / w_norm);
+
+      // # 2 MULTs
+      p.curr_vel_x = p.w_x * correction_factor;
+      p.curr_vel_y = p.w_y * correction_factor;
+
+      // # 2 MULTs, 2 ADDs
+      p.curr_pos_x = p.curr_pos_x + p.curr_vel_x * dt;
+      p.curr_pos_y = p.curr_pos_y + p.curr_vel_y * dt;
+
+      if (p.color == 0 && p.curr_pos_x > BOARD_LENGTH_MINUS_ONE_POINT_FIVE) {
+        p.curr_pos_x = 1.0;
+        p.destination_y = p.curr_pos_y;
+        p.w_x = p.desired_v0;
+        p.w_y = 0.0;
+        p.curr_vel_x = p.w_x;
+        p.curr_vel_y = p.w_y;
+      }
+      if (p.color == 1 && p.curr_pos_x < 1.5) {
+        p.curr_pos_x = BOARD_LENGTH_MINUS_ONE;
+        p.destination_y = p.curr_pos_y;
+        p.w_x = -p.desired_v0;
+        p.w_y = 0.0;
+        p.curr_vel_x = p.w_x;
+        p.curr_vel_y = p.w_y;
+      }
+
+      p.curr_pos_y = clamp(p.curr_pos_y, WALL_EPS, BOARD_WIDTH_MINUS_WALL_EPS);
+
+      // # 2 ADDs
+      double e_x = p.destination_x - p.curr_pos_x;
+      double e_y = p.destination_y - p.curr_pos_y;
+
+      // # 1 DIV, 1 SQRT, 2 MULTs, 1 ADD
+      double e_norm_inv = 1.0 / sqrt(e_x * e_x + e_y * e_y);
+
+      // # 2 MULTs
+      p.e_x = e_x * e_norm_inv;
+      p.e_y = e_y * e_norm_inv;
+
+      // # 2 MULTs, 1 ADD
+      double velocity_norm_sq = p.curr_vel_x * p.curr_vel_x + p.curr_vel_y * p.curr_vel_y;
+
+      // # 1 SQRT
+      double velocity_norm = sqrt(velocity_norm_sq);
+
+      // # 5 MULTs
+      p.s_b_sq = velocity_norm_sq * delta_t_sq;
+      p.s_b_e_x = p.e_x * (velocity_norm * delta_t);
+      p.s_b_e_y = p.e_y * (velocity_norm * delta_t);
+
+#if !BENCHMARK
+      if (dump_path && (t % dump_interval == 0)) {
+        samples.emplace_back(t + 1, i, p.curr_pos_x, p.curr_pos_y, p.e_x, p.e_y, p.color);
+      }
+#endif
+    }
+  }
+
+#if !BENCHMARK
+  if (dump_path) {
+    ofstream out(dump_path);
+    out << "time,id,x,y,p.e.x,p.e.y,color\n";
+
+    for (auto &[t, id, x, y, e_x, e_y, c] : samples) {
+      out << t << ',' << id << ',' << x << ',' << y << ',' << e_x << ',' << e_y << ',' << c << '\n';
+    }
+  }
+#endif
+
+  return 0;
+}