#ifndef GAIA_SIMD_NEIGHBOR_HPP
#define GAIA_SIMD_NEIGHBOR_HPP

#include <vector>
#include <cmath>
#include <algorithm>
#include <iostream>

#ifdef __AVX2__
#include <immintrin.h>
#endif

namespace gaia {
namespace simd {

struct Vec3 {
    float x, y, z;
    Vec3(float x=0, float y=0, float z=0) : x(x), y(y), z(z) {}
};

class SIMDNeighborList {
public:
    SIMDNeighborList(float cutoff = 10.0f) : cutoff(cutoff) {
        simd_width = 8;
    }
    
    void build(const std::vector<Vec3>& positions) {
        neighbors.clear();
        neighbors.resize(positions.size());
        
        #ifdef __AVX2__
        for (int i = 0; i < (int)positions.size(); i++) {
            float xi = positions[i].x;
            float yi = positions[i].y;
            float zi = positions[i].z;
            
            __m256 xi_vec = _mm256_set1_ps(xi);
            __m256 yi_vec = _mm256_set1_ps(yi);
            __m256 zi_vec = _mm256_set1_ps(zi);
            
            __m256 cutoff_sq = _mm256_set1_ps(cutoff * cutoff);
            
            for (int j = i + 1; j < (int)positions.size(); j += simd_width) {
                int remaining = std::min(simd_width, (int)positions.size() - j);
                
                float xj_arr[8] = {0}, yj_arr[8] = {0}, zj_arr[8] = {0};
                for (int k = 0; k < remaining; k++) {
                    xj_arr[k] = positions[j + k].x;
                    yj_arr[k] = positions[j + k].y;
                    zj_arr[k] = positions[j + k].z;
                }
                
                __m256 xj_vec = _mm256_loadu_ps(xj_arr);
                __m256 yj_vec = _mm256_loadu_ps(yj_arr);
                __m256 zj_vec = _mm256_loadu_ps(zj_arr);
                
                __m256 dx = _mm256_sub_ps(xi_vec, xj_vec);
                __m256 dy = _mm256_sub_ps(yi_vec, yj_vec);
                __m256 dz = _mm256_sub_ps(zi_vec, zj_vec);
                
                __m256 dx2 = _mm256_mul_ps(dx, dx);
                __m256 dy2 = _mm256_mul_ps(dy, dy);
                __m256 dz2 = _mm256_mul_ps(dz, dz);
                
                __m256 r2 = _mm256_add_ps(_mm256_add_ps(dx2, dy2), dz2);
                __m256 mask = _mm256_cmp_ps(r2, cutoff_sq, _CMP_LT_OQ);
                int mask_int = _mm256_movemask_ps(mask);
                
                for (int k = 0; k < remaining; k++) {
                    if (mask_int & (1 << k)) {
                        int idx = j + k;
                        if (idx < (int)positions.size()) {
                            neighbors[i].push_back(idx);
                            neighbors[idx].push_back(i);
                        }
                    }
                }
            }
        }
        #else
        for (int i = 0; i < (int)positions.size(); i++) {
            for (int j = i + 1; j < (int)positions.size(); j++) {
                float dx = positions[i].x - positions[j].x;
                float dy = positions[i].y - positions[j].y;
                float dz = positions[i].z - positions[j].z;
                float r2 = dx*dx + dy*dy + dz*dz;
                if (r2 < cutoff * cutoff) {
                    neighbors[i].push_back(j);
                    neighbors[j].push_back(i);
                }
            }
        }
        #endif
    }
    
    const std::vector<int>& get_neighbors(int i) const {
        return neighbors[i];
    }
    
    size_t total_pairs() const {
        size_t sum = 0;
        for (const auto& n : neighbors) sum += n.size();
        return sum / 2;
    }

private:
    float cutoff;
    int simd_width;
    std::vector<std::vector<int>> neighbors;
};

} // namespace simd
} // namespace gaia

#endif