/*
 *  This file is part of RawTherapee.
 *
 *  Copyright (c) 2019 Alberto Romei <aldrop8@gmail.com>
 *
 *  RawTherapee is free software: you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation, either version 3 of the License, or
 *  (at your option) any later version.
 *
 *  RawTherapee is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with RawTherapee.  If not, see <https://www.gnu.org/licenses/>.
 */
#include <cmath>
#include <iostream>

#include "rawimage.h"
#include "rawimagesource.h"
#include "improcfun.h"
#include "improccoordinator.h"
#include "imagefloat.h"

#include "coord.h"
#include "opthelper.h"
#include "pixelsmap.h"
#include "procparams.h"
#include "rt_algo.h"
#include "rtengine.h"
#include "rtthumbnail.h"
#include "sleef.h"
//#define BENCHMARK
//#include "StopWatch.h"
#include "iccstore.h"
#include "rt_math.h"
#include "color.h"


namespace
{
using rtengine::settings;
using rtengine::Coord2D;
using rtengine::ColorTemp;
using rtengine::findMinMaxPercentile;
using rtengine::ImageSource;
using rtengine::Imagefloat;
using rtengine::procparams::FilmNegativeParams;
using rtengine::procparams::ColorManagementParams;
using rtengine::procparams::RAWParams;
using rtengine::ICCStore;
using rtengine::MAXVALF;
using rtengine::CLIP;
using rtengine::TMatrix;
using rtengine::Color;
using RGB = rtengine::procparams::FilmNegativeParams::RGB;

Coord2D translateCoord(const rtengine::ImProcFunctions& ipf, int fw, int fh, int x, int y)
{

    const std::vector<Coord2D> points = {Coord2D(x, y)};

    std::vector<Coord2D> red;
    std::vector<Coord2D> green;
    std::vector<Coord2D> blue;
    ipf.transCoord(fw, fh, points, red, green, blue);

    return green[0];
}


void getSpotAvgMax(ImageSource *imgsrc, ColorTemp currWB, const std::unique_ptr<rtengine::procparams::ProcParams> &params,
                   Coord2D p, int tr, int spotSize, RGB &avg, RGB &max)
{
    int x1 = MAX(0, (int)p.x - spotSize / 2);
    int y1 = MAX(0, (int)p.y - spotSize / 2);
    PreviewProps pp(x1, y1, spotSize, spotSize, 1);

    if (settings->verbose) {
        printf("Spot: %d,%d   %d,%d\n", x1, y1, x1 + spotSize / 2, y1 + spotSize / 2);
    }

    rtengine::Imagefloat spotImg(spotSize, spotSize);
    imgsrc->getImage(currWB, tr, &spotImg, pp, params->toneCurve, params->raw);

    auto avgMax = [spotSize, &spotImg](RGB & avg, RGB & max) -> void {
        avg = {};
        max = {};

        for (int i = 0; i < spotSize; ++i)
        {
            for (int j = 0; j < spotSize; ++j) {

                float r = spotImg.r(i, j);
                float g = spotImg.g(i, j);
                float b = spotImg.b(i, j);

                avg.r += r;
                avg.g += g;
                avg.b += b;
                max.r = MAX(max.r, r);
                max.g = MAX(max.g, g);
                max.b = MAX(max.b, b);
            }
        }

        avg.r /= (spotSize * spotSize);
        avg.g /= (spotSize * spotSize);
        avg.b /= (spotSize * spotSize);
    };

    if (params->filmNegative.colorSpace == rtengine::FilmNegativeParams::ColorSpace::INPUT) {
        avgMax(avg, max);
    } else {
        // Convert spot image to current working space
        imgsrc->convertColorSpace(&spotImg, params->icm, currWB);

        avgMax(avg, max);

        // TODO handle custom color profile !
    }

    // Clip away zeroes or negative numbers, you never know
    avg.r = MAX(avg.r, 1.f);
    avg.g = MAX(avg.g, 1.f);
    avg.b = MAX(avg.b, 1.f);

    if (settings->verbose) {
        printf("Average Spot RGB: %f,%f,%f\n", avg.r, avg.g, avg.b);
    }
}


void calcMedians(
    const rtengine::Imagefloat* input,
    int x1, int y1, int x2, int y2,
    float &rmed, float &gmed, float &bmed
)
{
    using rtengine::findMinMaxPercentile;

    // Channel vectors to calculate medians
    std::vector<float> rv, gv, bv;

    const int sz = (x2 - x1) * (y2 - y1);
    rv.reserve(sz);
    gv.reserve(sz);
    bv.reserve(sz);


    for (int ii = y1; ii < y2; ii ++) {
        for (int jj = x1; jj < x2; jj ++) {
            rv.push_back(input->r(ii, jj));
            gv.push_back(input->g(ii, jj));
            bv.push_back(input->b(ii, jj));
        }
    }

    // Calculate channel medians of the specified area
    findMinMaxPercentile(rv.data(), rv.size(), 0.5f, rmed, 0.5f, rmed, true);
    findMinMaxPercentile(gv.data(), gv.size(), 0.5f, gmed, 0.5f, gmed, true);
    findMinMaxPercentile(bv.data(), bv.size(), 0.5f, bmed, 0.5f, bmed, true);
}


RGB getMedians(const rtengine::Imagefloat* input, int borderPercent)
{
    float rmed, gmed, bmed;
    // Cut 20% border from medians calculation. It will probably contain outlier values
    // from the film holder, which will bias the median result.
    const int bW = input->getWidth() * borderPercent / 100;
    const int bH = input->getHeight() * borderPercent / 100;
    calcMedians(input, bW, bH,
                input->getWidth() - bW, input->getHeight() - bH,
                rmed, gmed, bmed);

    if (settings->verbose) {
        printf("Channel medians: R=%g, G=%g, B=%g\n", rmed, gmed, bmed);
    }

    return { rmed, gmed, bmed };
}

/*
// TODO not needed for now
void convertColorSpace(Imagefloat* input, const TMatrix &src2xyz, const TMatrix &xyz2dest)
{

#ifdef _OPENMP
    #pragma omp parallel for
#endif

    for (int i = 0; i < input->getHeight(); i++) {
        for (int j = 0; j < input->getWidth(); j++) {

            float newr = input->r(i, j);
            float newg = input->g(i, j);
            float newb = input->b(i, j);

            float x, y, z;
            Color::rgbxyz (newr, newg, newb, x, y, z, src2xyz);
            Color::xyz2rgb (x, y, z, newr, newg, newb, xyz2dest);

            input->r(i, j) = newr;
            input->g(i, j) = newg;
            input->b(i, j) = newb;

        }
    }
}
*/


/**
 * Perform actual film negative inversion process.
 * Returns true if the input and output reference values are not set in params; refIn/refOut will be updated with median-based estimates.
 * Otherwise, use provided values in params and return false
 */
bool doProcess(Imagefloat *input, Imagefloat *output,
               const FilmNegativeParams &params, const ColorManagementParams &icmParams,
               RGB &refIn, RGB &refOut)
{
    bool refsUpdated = false;

    float rexp = -(params.greenExp * params.redRatio);
    float gexp = -params.greenExp;
    float bexp = -(params.greenExp * params.blueRatio);

    // In case we are processing a thumbnail, reference values might not be set in params,
    // so make an estimate on the fly, using channel medians
    if (refIn.g <= 0.f) {
        // Calc medians, 20% border cut
        refIn = getMedians(input, 20);
        refsUpdated = true;
    } else {
        refIn = params.refInput;
    }

    if (refOut.g <= 0.f) {
        // Median will correspond to gray, 1/24th of max in the output
        refOut = { MAXVALF / 24.f, MAXVALF / 24.f, MAXVALF / 24.f };
        refsUpdated = true;
    } else {
        refOut = params.refOutput;
    }

    // Apply channel exponents to reference input values, and compute suitable multipliers
    // in order to reach reference output values.
    float rmult = refOut.r / pow_F(rtengine::max(refIn.r, 1.f), rexp);
    float gmult = refOut.g / pow_F(rtengine::max(refIn.g, 1.f), gexp);
    float bmult = refOut.b / pow_F(rtengine::max(refIn.b, 1.f), bexp);


#ifdef __SSE2__
    const vfloat clipv = F2V(MAXVALF);
    const vfloat rexpv = F2V(rexp);
    const vfloat gexpv = F2V(gexp);
    const vfloat bexpv = F2V(bexp);
    const vfloat rmultv = F2V(rmult);
    const vfloat gmultv = F2V(gmult);
    const vfloat bmultv = F2V(bmult);
#endif

    const int rheight = input->getHeight();
    const int rwidth = input->getWidth();

    for (int i = 0; i < rheight; i++) {
        float *rlinein = input->r(i);
        float *glinein = input->g(i);
        float *blinein = input->b(i);
        float *rlineout = output->r(i);
        float *glineout = output->g(i);
        float *blineout = output->b(i);
        int j = 0;
#ifdef __SSE2__

        for (; j < rwidth - 3; j += 4) {
            STVFU(rlineout[j], vminf(rmultv * pow_F(LVFU(rlinein[j]), rexpv), clipv));
            STVFU(glineout[j], vminf(gmultv * pow_F(LVFU(glinein[j]), gexpv), clipv));
            STVFU(blineout[j], vminf(bmultv * pow_F(LVFU(blinein[j]), bexpv), clipv));
        }

#endif

        for (; j < rwidth; ++j) {
            rlineout[j] = CLIP(rmult * pow_F(rlinein[j], rexp));
            glineout[j] = CLIP(gmult * pow_F(glinein[j], gexp));
            blineout[j] = CLIP(bmult * pow_F(blinein[j], bexp));
        }
    }

    return refsUpdated;
}


}



bool rtengine::ImProcFunctions::filmNegativeProcess(
    Imagefloat *input, Imagefloat *output, FilmNegativeParams &fnp,
    const RAWParams &rawParams, const ImageSource* imgsrc, const ColorTemp &currWB)
{
    //BENCHFUNMICRO

    if (!fnp.enabled) {
        return false;
    }

    bool paramsUpdated = false;

    RGB &refIn = fnp.refInput;
    RGB &refOut = fnp.refOutput;

    // If we're opening a profile from an older version, apply the proper multiplier
    // compensations to make processing backwards compatible.

    if (fnp.backCompat == FilmNegativeParams::BackCompat::V1) {
        // Calc medians, no border cut, compensate currWB in+out
        refIn = getMedians(input, 0);
        refOut = { MAXVALF / 24.f, MAXVALF / 24.f, MAXVALF / 24.f };

        std::array<float, 4> scale_mul = { 1.f, 1.f, 1.f, 1.f };
        float autoGainComp, rm, gm, bm;
        imgsrc->getWBMults(currWB, params->raw, scale_mul, autoGainComp, rm, gm, bm);

        refOut.r *= rm;
        refOut.g *= gm;
        refOut.b *= bm;

        paramsUpdated = true;

    } else if (fnp.backCompat == FilmNegativeParams::BackCompat::V2) {

        std::array<float, 4> scale_mul = { 1.f, 1.f, 1.f, 1.f };
        float autoGainComp, rm, gm, bm;
        imgsrc->getWBMults(currWB, params->raw, scale_mul, autoGainComp, rm, gm, bm);

        float rm2, gm2, bm2;
        imgsrc->getWBMults(rtengine::ColorTemp(3500., 1., 1., "Custom", currWB.getObserver()), params->raw, scale_mul, autoGainComp, rm2, gm2, bm2);
        float mg = rtengine::max(rm2, gm2, bm2);
        rm2 /= mg;
        gm2 /= mg;
        bm2 /= mg;

        if (fnp.refInput.g == 0.f) {
            // Calc medians, 20% border cut
            refIn = getMedians(input, 20);
            refOut = { MAXVALF / 24.f, MAXVALF / 24.f, MAXVALF / 24.f };
        } else if (fnp.refInput.g > 0.f) {
            // Calc refInput + refOutput from base levels, compensate currWB in, 3500 out
            refIn = fnp.refInput;
            refIn.r *= rm * scale_mul[0];
            refIn.g *= gm * scale_mul[1];
            refIn.b *= bm * scale_mul[2];
            refOut = { MAXVALF / 512.f, MAXVALF / 512.f, MAXVALF / 512.f };
        }

        refOut.r *= rm * autoGainComp / rm2;
        refOut.g *= gm * autoGainComp / gm2;
        refOut.b *= bm * autoGainComp / bm2;

        paramsUpdated = true;

    }

    if (settings->verbose && fnp.backCompat != FilmNegativeParams::BackCompat::CURRENT) {
        printf("Upgraded from V%d - refIn: R=%g G=%g B=%g refOut: R=%g G=%g B=%g\n",
               (int)fnp.backCompat,
               static_cast<double>(refIn.r), static_cast<double>(refIn.g), static_cast<double>(refIn.b),
               static_cast<double>(refOut.r), static_cast<double>(refOut.g), static_cast<double>(refOut.b));
    }

    // FilmNeg params are now upgraded to the latest version
    fnp.backCompat = FilmNegativeParams::BackCompat::CURRENT;

    // Perform actual inversion. Return true if reference values are computed from medians
    paramsUpdated |= doProcess(input, output, fnp, this->params->icm, refIn, refOut);

    return paramsUpdated;

}

void rtengine::ImProcFunctions::filmNegativeProcess(rtengine::Imagefloat *input, rtengine::Imagefloat *output,
        const procparams::FilmNegativeParams &params)
{
    //BENCHFUNMICRO

    if (!params.enabled) {
        return;
    }

    RGB refIn = params.refInput, refOut = params.refOutput;

    doProcess(input, output, params, this->params->icm, refIn, refOut);

}


bool rtengine::ImProcCoordinator::getFilmNegativeSpot(int x, int y, const int spotSize, RGB &refInput, RGB &refOutput)
{
    MyMutex::MyLock lock(mProcessing);

    const int tr = getCoarseBitMask(params->coarse);

    const Coord2D p = translateCoord(ipf, fw, fh, x, y);

    // Get the average channel values from the sampled spot
    RGB avg, max;
    getSpotAvgMax(imgsrc, currWB, params, p, tr, spotSize, avg, max);

    float rexp = -(params->filmNegative.greenExp * params->filmNegative.redRatio);
    float gexp = -params->filmNegative.greenExp;
    float bexp = -(params->filmNegative.greenExp * params->filmNegative.blueRatio);

    float rmult = params->filmNegative.refOutput.r / pow_F(rtengine::max(params->filmNegative.refInput.r, 1.f), rexp);
    float gmult = params->filmNegative.refOutput.g / pow_F(rtengine::max(params->filmNegative.refInput.g, 1.f), gexp);
    float bmult = params->filmNegative.refOutput.b / pow_F(rtengine::max(params->filmNegative.refInput.b, 1.f), bexp);

    refInput = avg;

    refOutput.r = rmult * pow_F(avg.r, rexp);
    refOutput.g = gmult * pow_F(avg.g, gexp);
    refOutput.b = bmult * pow_F(avg.b, bexp);

    return true;
}







// ---------- >>> legacy mode >>> ---------------


// For backwards compatibility with profiles saved by RT 5.7 - 5.8
void rtengine::Thumbnail::processFilmNegative(
    const procparams::ProcParams &params,
    const Imagefloat* baseImg,
    const int rwidth, const int rheight
)
{

    // Channel exponents
    const float rexp = -params.filmNegative.redRatio * params.filmNegative.greenExp;
    const float gexp = -params.filmNegative.greenExp;
    const float bexp = -params.filmNegative.blueRatio * params.filmNegative.greenExp;

    const float MAX_OUT_VALUE = 65000.f;

    // Channel medians
    float rmed, gmed, bmed;

    // If using the old method, calculate medians on the whole image
    calcMedians(baseImg, 0, 0, rwidth, rheight, rmed, gmed, bmed);

    if (settings->verbose) {
        printf("FilmNeg legacy V1 :: Thumbnail input channel medians: %g %g %g\n", rmed, gmed, bmed);
    }

    // Calculate output medians
    rmed = powf(rmed, rexp);
    gmed = powf(gmed, gexp);
    bmed = powf(bmed, bexp);

    // Calculate output multipliers so that the median value is 1/24 of the output range.
    float rmult, gmult, bmult;
    rmult = (MAX_OUT_VALUE / 24.f) / rmed;
    gmult = (MAX_OUT_VALUE / 24.f) / gmed;
    bmult = (MAX_OUT_VALUE / 24.f) / bmed;

    float rsum = 0.f, gsum = 0.f, bsum = 0.f;

    for (int i = 0; i < rheight; i++) {
        for (int j = 0; j < rwidth; j++) {
            rsum += baseImg->r(i, j);
            gsum += baseImg->g(i, j);
            bsum += baseImg->b(i, j);
        }
    }

    const float ravg = rsum / (rheight * rwidth);
    const float gavg = gsum / (rheight * rwidth);
    const float bavg = bsum / (rheight * rwidth);

    // Shifting current WB multipliers, based on channel averages.
    rmult /= gavg / ravg;
    // gmult /= gAvg / gAvg;  green chosen as reference channel
    bmult /= gavg / bavg;

    if (settings->verbose) {
        printf("FilmNeg legacy V1 :: Thumbnail computed multipliers: %g %g %g\n", static_cast<double>(rmult), static_cast<double>(gmult), static_cast<double>(bmult));
    }

#ifdef __SSE2__
    const vfloat clipv = F2V(MAXVALF);
    const vfloat rexpv = F2V(rexp);
    const vfloat gexpv = F2V(gexp);
    const vfloat bexpv = F2V(bexp);
    const vfloat rmultv = F2V(rmult);
    const vfloat gmultv = F2V(gmult);
    const vfloat bmultv = F2V(bmult);
#endif

    for (int i = 0; i < rheight; i++) {
        float *rline = baseImg->r(i);
        float *gline = baseImg->g(i);
        float *bline = baseImg->b(i);
        int j = 0;
#ifdef __SSE2__

        for (; j < rwidth - 3; j += 4) {
            STVFU(rline[j], vminf(rmultv * pow_F(LVFU(rline[j]), rexpv), clipv));
            STVFU(gline[j], vminf(gmultv * pow_F(LVFU(gline[j]), gexpv), clipv));
            STVFU(bline[j], vminf(bmultv * pow_F(LVFU(bline[j]), bexpv), clipv));
        }

#endif

        for (; j < rwidth; ++j) {
            rline[j] = CLIP(rmult * pow_F(rline[j], rexp));
            gline[j] = CLIP(gmult * pow_F(gline[j], gexp));
            bline[j] = CLIP(bmult * pow_F(bline[j], bexp));
        }
    }
}


// For backwards compatibility with intermediate dev version (see filmneg_stable_mults branch)
void rtengine::Thumbnail::processFilmNegativeV2(
    const procparams::ProcParams &params,
    const Imagefloat* baseImg,
    const int rwidth, const int rheight
)
{

    // Channel exponents
    const float rexp = -params.filmNegative.redRatio * params.filmNegative.greenExp;
    const float gexp = -params.filmNegative.greenExp;
    const float bexp = -params.filmNegative.blueRatio * params.filmNegative.greenExp;

    // Calculate output multipliers
    float rmult, gmult, bmult;

    const float MAX_OUT_VALUE = 65000.f;

    // If the film base values are not set in params, estimate multipliers from each channel's median value.
    if (params.filmNegative.refInput.r <= 0.f) {

        // Channel medians
        float rmed, gmed, bmed;

        // The new method cuts out a 20% border from medians calculation.
        const int bW = rwidth * 20 / 100;
        const int bH = rheight * 20 / 100;
        calcMedians(baseImg, bW, bH, rwidth - bW, rheight - bH, rmed, gmed, bmed);

        if (settings->verbose) {
            printf("FilmNeg legacy V2 :: Thumbnail input channel medians: %g %g %g\n", rmed, gmed, bmed);
        }

        // Calculate output medians
        rmed = powf(rmed, rexp);
        gmed = powf(gmed, gexp);
        bmed = powf(bmed, bexp);

        // Calculate output multipliers so that the median value is 1/24 of the output range.
        rmult = (MAX_OUT_VALUE / 24.f) / rmed;
        gmult = (MAX_OUT_VALUE / 24.f) / gmed;
        bmult = (MAX_OUT_VALUE / 24.f) / bmed;

    } else {

        // Read film-base values from params
        float rbase = params.filmNegative.refInput.r; // redBase;
        float gbase = params.filmNegative.refInput.g; // greenBase;
        float bbase = params.filmNegative.refInput.b; // blueBase;

        // Reconstruct scale_mul coefficients from thumbnail metadata:
        //   redMultiplier / camwbRed is pre_mul[0]
        //   pre_mul[0] * scaleGain is scale_mul[0]
        // Apply channel scaling to raw (unscaled) input base values, to
        // match with actual (scaled) data in baseImg
        rbase *= (redMultiplier / camwbRed)     * scaleGain;
        gbase *= (greenMultiplier / camwbGreen) * scaleGain;
        bbase *= (blueMultiplier / camwbBlue)   * scaleGain;

        if (settings->verbose) {
            printf("FilmNeg legacy V2 :: Thumbnail input film base values: %g %g %g\n", rbase, gbase, bbase);
        }

        // Apply exponents to get output film base values
        rbase = powf(rbase, rexp);
        gbase = powf(gbase, gexp);
        bbase = powf(bbase, bexp);

        // Calculate multipliers so that film base value is 1/512th of the output range.
        rmult = (MAX_OUT_VALUE / 512.f) / rbase;
        gmult = (MAX_OUT_VALUE / 512.f) / gbase;
        bmult = (MAX_OUT_VALUE / 512.f) / bbase;

    }


    // Get and un-apply multipliers to adapt the thumbnail to a known fixed WB setting,
    // as in the main image processing.

    double r, g, b;
    ColorTemp(3500., 1., 1., "Custom", params.wb.observer).getMultipliers(r, g, b);
    //iColorMatrix is cam_rgb
    const double rm = camwbRed   / (iColorMatrix[0][0] * r + iColorMatrix[0][1] * g + iColorMatrix[0][2] * b);
    const double gm = camwbGreen / (iColorMatrix[1][0] * r + iColorMatrix[1][1] * g + iColorMatrix[1][2] * b);
    const double bm = camwbBlue  / (iColorMatrix[2][0] * r + iColorMatrix[2][1] * g + iColorMatrix[2][2] * b);

    // Normalize max WB multiplier to 1.f
    const double m = max(rm, gm, bm);
    rmult /= rm / m;
    gmult /= gm / m;
    bmult /= bm / m;


    if (settings->verbose) {
        printf("FilmNeg legacy V2 :: Thumbnail computed multipliers: %g %g %g\n", static_cast<double>(rmult), static_cast<double>(gmult), static_cast<double>(bmult));
    }


#ifdef __SSE2__
    const vfloat clipv = F2V(MAXVALF);
    const vfloat rexpv = F2V(rexp);
    const vfloat gexpv = F2V(gexp);
    const vfloat bexpv = F2V(bexp);
    const vfloat rmultv = F2V(rmult);
    const vfloat gmultv = F2V(gmult);
    const vfloat bmultv = F2V(bmult);
#endif

    for (int i = 0; i < rheight; i++) {
        float *rline = baseImg->r(i);
        float *gline = baseImg->g(i);
        float *bline = baseImg->b(i);
        int j = 0;
#ifdef __SSE2__

        for (; j < rwidth - 3; j += 4) {
            STVFU(rline[j], vminf(rmultv * pow_F(LVFU(rline[j]), rexpv), clipv));
            STVFU(gline[j], vminf(gmultv * pow_F(LVFU(gline[j]), gexpv), clipv));
            STVFU(bline[j], vminf(bmultv * pow_F(LVFU(bline[j]), bexpv), clipv));
        }

#endif

        for (; j < rwidth; ++j) {
            rline[j] = CLIP(rmult * pow_F(rline[j], rexp));
            gline[j] = CLIP(gmult * pow_F(gline[j], gexp));
            bline[j] = CLIP(bmult * pow_F(bline[j], bexp));
        }
    }
}

// ----------------- <<< legacy mode <<< ------------