// Copyright (2019-2024) Paul Scherrer Institute

#include "NeuralNetResPredictor.h"

#include <cmath>
#include "../common/JFJochException.h"

NeuralNetResPredictor::NeuralNetResPredictor(const std::string& model_path)
        : model_input(512*512),
          enable(!model_path.empty())
#ifdef JFJOCH_USE_TORCH
        , device(torch::kCUDA)
#endif
{
#ifdef JFJOCH_USE_TORCH
    if (enable) {
        module = torch::jit::load(model_path);
        module.to(device);
    }
#else
    enable = false;
#endif
}

template<class T>
void NeuralNetResPredictor::PrepareInternal(const DiffractionExperiment& experiment, const T* image) {
    size_t pool_factor = GetMaxPoolFactor(experiment);
    size_t xpixel = experiment.GetXPixelsNum();
    size_t ypixel = experiment.GetYPixelsNum();
    size_t start_x = std::lround(experiment.GetBeamX_pxl());
    size_t start_y = std::lround(experiment.GetBeamY_pxl());

    for (size_t y = 0; y < 512; y++) {
        size_t y0 = y * pool_factor + start_y;
        size_t min_yp = std::min(y0 + pool_factor, ypixel);
        for (size_t x = 0; x < 512; x++) {
            float val = 0.0;
            size_t x0 = x * pool_factor + start_x;
            size_t min_xp = std::min(x0 + pool_factor, ypixel);

            for (size_t yp = y0; yp < min_yp; yp++) {
                for (size_t xp = x0; xp < min_xp; xp++) {
                    int16_t pxl = image[yp * xpixel + xp];
                    if (val < pxl)
                        val = pxl;
                }
            }
            float max_pool = floorf(sqrtf(val));
            model_input[512 * y + x] = max_pool;
        }
    }
}

void NeuralNetResPredictor::Prepare(const DiffractionExperiment& experiment, const int16_t *image) {
    PrepareInternal(experiment, image);
}

void NeuralNetResPredictor::Prepare(const DiffractionExperiment& experiment, const int32_t *image) {
    PrepareInternal(experiment, image);
}

size_t NeuralNetResPredictor::GetMaxPoolFactor(const DiffractionExperiment& experiment) const {
    float max_direction = std::max(experiment.GetXPixelsNum(), experiment.GetYPixelsNum()) / 2.0;
    size_t pool_factor = std::lround(max_direction / 512.0f);
    if (pool_factor <= 0)
        throw JFJochException(JFJochExceptionCategory::InputParameterInvalid, "Detector size is too small");

    if (pool_factor > 8)
        throw JFJochException(JFJochExceptionCategory::InputParameterInvalid, "Detector size is too large");
    return pool_factor;
}

std::optional<float> NeuralNetResPredictor::Inference(const DiffractionExperiment& experiment, const void *image) {
    if (!enable)
        return {};
#ifdef JFJOCH_USE_TORCH
    if (experiment.GetPixelDepth() == 2)
        Prepare(experiment, (int16_t *) image);
    else
        Prepare(experiment, (int32_t *) image);

    auto options = torch::TensorOptions().dtype(at::kFloat);
    auto model_input_tensor = torch::from_blob(model_input.data(), {1,1,512,512}, options).to(device);

    std::vector<torch::jit::IValue> pixels;
    pixels.emplace_back(std::move(model_input_tensor));

    auto output = module.forward(pixels).toTensor();
    auto tensor_output = output[0].item<float>();
    float two_theta = atanf(((2.0f * experiment.GetPixelSize_mm() / experiment.GetDetectorDistance_mm()) * tensor_output));
    float stheta = sinf(two_theta * 0.5f);
    float resolution = experiment.GetWavelength_A() / (2.0f * stheta);
    return resolution;
#else
    return {};
#endif
}

const std::vector<float> &NeuralNetResPredictor::GetModelInput() const {
    return model_input;
}