#include "framed_model.hpp"
#include "twin.hpp"
#include "freqpath.hpp"
#include <cmath>
#include <cstring>
#include <algorithm>

namespace {

constexpr float G_FIT = 0.9963f;
constexpr float W_FIT = 0.3335f;
constexpr float A_FIT = 0.9807f;
constexpr float GAIN = 4.132f;              // sqrtf(param_5), sens_lin = GAIN^2

// framed_render.py LUT nodes (Pchip): C = G_FIT*LUT(xv) + W_FIT*warp^A_FIT
constexpr double LX[12] = {-0.75, -0.5012, -0.5, -0.2012, 0.0988, 0.2488,
                           0.3988, 0.5488, 0.574, 0.61, 0.75, 1.0};
constexpr double LY[12] = {0.4402, 0.366, 0.4552, 0.459, 0.541, 0.576,
                           0.608, 0.636, 0.5645, 0.6471, 0.6562, 0.6670};
constexpr int LN = 12;

// PchipInterpolator (monotone cubic Hermite, scipy semantics) eval at one x.
double pchip_eval(double x) {
    if (x <= LX[0]) return LY[0];
    if (x >= LX[LN - 1]) return LY[LN - 1];
    // slopes
    double m[LN];
    for (int i = 0; i < LN - 1; i++) m[i] = (LY[i + 1] - LY[i]) / (LX[i + 1] - LX[i]);
    double d[LN];
    d[0] = m[0];
    for (int i = 1; i < LN - 1; i++) {
        if (m[i - 1] * m[i] <= 0.0) d[i] = 0.0;
        else {
            double w1 = 2 * LX[i + 1] - LX[i] - LX[i - 1];
            double w2 = LX[i + 1] - LX[i - 1];
            d[i] = (w1 + w2) / ((w1 / m[i - 1]) + (w2 / m[i]));
        }
    }
    d[LN - 1] = m[LN - 2];
    int i = 0;
    while (i < LN - 2 && x > LX[i + 1]) i++;
    double h = LX[i + 1] - LX[i];
    double t = (x - LX[i]) / h;
    double y0 = LY[i], y1 = LY[i + 1], d0 = d[i], d1 = d[i + 1];
    double h00 = 2 * t * t * t - 3 * t * t + 1;
    double h10 = t * t * t - 2 * t * t + t;
    double h01 = -2 * t * t * t + 3 * t * t;
    double h11 = t * t * t - t * t;
    double v = h00 * y0 + h10 * h * d0 + h01 * y1 + h11 * h * d1;
    return std::max(LY[LN - 2] < LY[LN - 1] ? 0.0 : -1e9,
                    std::min(v, *std::max_element(LY, LY + LN) * 1.0));
}

double lut_eval(double x) {
    double v = pchip_eval(x);
    double lo = *std::min_element(LY, LY + LN);
    double hi = *std::max_element(LY, LY + LN);
    return std::max(lo, std::min(hi, v));
}

}  // namespace

FramedDetector::FramedDetector(size_t nfft, float sample_rate)
    : nfft_(nfft), sample_rate_(sample_rate), fc_(0), q_(1.0), wsum_(0), inited_(false) {
    res_.resize(nfft / 2 + 1, 1.0f);
    warp_.resize(nfft, 0.0f);
    am_.resize(nfft / 2 + 1, 0.0f);
}

FramedDetector::~FramedDetector() {}

void FramedDetector::setParams(float fc, float q) {
    fc_ = fc;
    q_ = std::max(1.0f, q);
    size_t half = nfft_ / 2;

    // twin resonance per bin (res = |2B(z)/A(z)|)
    float sens_lin = GAIN * GAIN;  // param_5
    detkernel::twin_coeff c = detkernel::build_twin_coeff(
        static_cast<double>(sample_rate_), static_cast<double>(fc),
        static_cast<double>(q_), sens_lin);
    std::vector<detkernel::cplxf> z(half + 1);
    std::vector<detkernel::cplxf> out(half + 1);
    for (size_t k = 0; k <= half; k++) {
        double theta = 2.0 * M_PI * static_cast<double>(k) / static_cast<double>(nfft_);
        z[k].re = static_cast<float>(std::cos(theta));
        z[k].im = static_cast<float>(std::sin(theta));
    }
    detkernel::twin_apply(c, z.data(), half + 1, out.data());
    for (size_t k = 0; k <= half; k++) {
        res_[k] = std::sqrt(out[k].re * out[k].re + out[k].im * out[k].im);
        res_[k] = std::max(res_[k], 1e-12f);
    }

    // warp (freqpath 0x5406a8)
    detkernel::build_warp(static_cast<float>(sample_rate_),
                          static_cast<int>(nfft_), warp_.data());

    inited_ = true;
}

void FramedDetector::processFrame(const std::complex<double>* spectrum, float* mask) {
    size_t half = nfft_ / 2;
    // wsum = sum of sqrt-hann window (compute once via nfft)
    if (wsum_ == 0.0) {
        double s = 0.0;
        for (size_t i = 0; i < nfft_; i++) {
            s += std::sqrt(0.5 * (1.0 - std::cos(2.0 * M_PI * i / (nfft_ - 1))));
        }
        wsum_ = s;
    }

    double tatt = 0.011, trel = 0.08;
    double att = std::exp(-1.0 * (nfft_ / 4) / (tatt * sample_rate_));
    double rel = std::exp(-1.0 * (nfft_ / 4) / (trel * sample_rate_));

    double rp = 0.0275 * std::pow(static_cast<double>(q_), 0.2159);

    for (size_t k = 0; k <= half; k++) {
        double a_cur = 2.0 * std::abs(spectrum[k]) / wsum_;
        double am = am_[k];
        if (a_cur > am) am = att * am + (1.0 - att) * a_cur;
        else am = rel * am + (1.0 - rel) * a_cur;
        am_[k] = static_cast<float>(am);

        double xv = std::log10(std::max(am / static_cast<double>(res_[k]), 1e-9));
        double C = G_FIT * lut_eval(xv) +
                   W_FIT * std::pow(static_cast<double>(warp_[k]), A_FIT);
        double gain = std::max(1.0 - C, 1e-9) *
                      std::pow(static_cast<double>(res_[k]), rp);
        mask[k] = static_cast<float>(gain);
    }
    for (size_t k = half + 1; k < nfft_; k++) {
        mask[k] = mask[nfft_ - k];
    }
}