// # 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 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); }

static inline __m512d _avx512_exp_pd(__m512d x) {

  const __m512d log2e = _mm512_set1_pd(1.4426950408889634);
  __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);
  __m512d r = _mm512_fnmadd_pd(nf, ln2, x);

  const __m512d c7 = _mm512_set1_pd(1.0 / 5040.0);
  const __m512d c6 = _mm512_set1_pd(1.0 / 720.0);
  const __m512d c5 = _mm512_set1_pd(1.0 / 120.0);
  const __m512d c4 = _mm512_set1_pd(1.0 / 24.0);
  const __m512d c3 = _mm512_set1_pd(1.0 / 6.0);
  const __m512d c2 = _mm512_set1_pd(0.5);
  const __m512d c1 = _mm512_set1_pd(1.0);
  const __m512d c0 = _mm512_set1_pd(1.0);

  __m512d p = 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);

  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);

  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 smc = _mm512_div_pd(numer, denom);

        __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_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]);

      __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));
      __m512d w_norm = _mm512_sqrt_pd(w_norm_sq);
      __m512d correction_factor = _mm512_min_pd(_mm512_set1_pd(1.0), _mm512_div_pd(v_max_i_vec, w_norm));

      __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);
      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 - 1.5), _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));

      __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);
      __m512d e_norm_sq = _mm512_fmadd_pd(ex_vec, ex_vec, _mm512_mul_pd(ey_vec, ey_vec));
      __m512d e_norm = _mm512_sqrt_pd(e_norm_sq);
      __m512d e_norm_inv = _mm512_div_pd(_mm512_set1_pd(1.0), e_norm);
      __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);

      __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));
      __m512d sb_sq_i_vec = _mm512_mul_pd(vel_norm_sq, _mm512_set1_pd(delta_t_sq));
      __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);
      _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;
}