// SPDX-FileCopyrightText: 2024 Filip Leonarski, Paul Scherrer Institute <filip.leonarski@psi.ch>
// SPDX-License-Identifier: GPL-3.0-only

#include <catch2/catch_all.hpp>

#include "../image_analysis/bragg_prediction/BraggPrediction.h"
#include <iostream>
#include "../image_analysis/SensorAbsorption.h"

// The flight-path term is a zero-parameter prediction: a NIST attenuation coefficient, the stated
// sample-to-detector distance, and Beer-Lambert. Nothing about it is fitted, so it can be checked
// against the tables it is computed from rather than against any measurement.
TEST_CASE("BraggPrediction_FlightPathAttenuation", "[air_path]") {
    using sensor_absorption::FlightPathAttenuation;
    constexpr double D_mm = 160.0;
    auto lambda_of = [](double keV) { return 12.39842 / keV; };
    auto air = [&](double keV) {
        return FlightPathAttenuation::Build(FlightPathMedium::Air, D_mm, lambda_of(keV));
    };

    // Attenuation length of dry air, as D/L at a known distance. These follow from the NIST dry-air
    // mass attenuation coefficients and 1.205e-3 g/cm^3, and span four decades over the corpus's
    // energy range: L is 8.2 m at 18 keV but only 8.9 cm at 3.8 keV.
    CHECK(air(18.0).d_over_L == Catch::Approx(0.01959).epsilon(0.01));
    CHECK(air(12.4).d_over_L == Catch::Approx(0.05351).epsilon(0.01));
    CHECK(air(3.76).d_over_L == Catch::Approx(1.7972).epsilon(0.01));

    // The argon K edge at 3.2029 keV is in the table and is a real step: argon is 1.28% of air by
    // mass but dominates its absorption here, so interpolating straight across it would understate
    // the correction in exactly the regime where the correction is largest.
    CHECK(air(3.21).d_over_L > air(3.19).d_over_L * 1.05f);

    const auto a18 = air(18.0);
    // Normalised at normal incidence: head-on is untouched, by construction rather than by rounding.
    CHECK(a18.Factor(1.0f) == 1.0f);
    // Always >= 1 and monotone in the obliquity, because a reflection that arrives at a larger angle
    // crossed strictly more of the medium.
    float prev = 1.0f;
    for (double alpha_deg : {10.0, 20.0, 30.0, 40.0, 50.0, 55.0}) {
        const float f = a18.Factor(static_cast<float>(std::cos(alpha_deg * M_PI / 180.0)));
        CHECK(f > prev);
        prev = f;
    }
    const float cos55 = static_cast<float>(std::cos(55.0 * M_PI / 180.0));
    // 18 keV over 160 mm at 55 degrees: +1.5%.
    CHECK(a18.Factor(cos55) == Catch::Approx(1.0147).epsilon(0.002));

    // It runs the other way to the sensor term at the same angle: the sensor makes an oblique
    // reflection read high, the medium makes it read low, and neither is the other's undoing.
    const auto qe = sensor_absorption::SensorQE::Build("Si", 450.0, lambda_of(18.0));
    CHECK(qe.Factor(cos55) < 1.0f);
    CHECK(a18.Factor(cos55) > 1.0f);

    // HELIUM is why a long-wavelength station is usable at all: at 3.8 keV it attenuates some three
    // orders of magnitude less than air, so the same geometry that costs a factor of several in air
    // costs a fraction of a per cent in helium. It is NOT vacuum, and is not modelled as one.
    const auto he = FlightPathAttenuation::Build(FlightPathMedium::Helium, D_mm, lambda_of(3.76));
    CHECK(he.d_over_L > 0.0f);
    CHECK(he.d_over_L < air(3.76).d_over_L / 100.0f);
    CHECK(he.Factor(cos55) > 1.0f);
    CHECK(he.Factor(cos55) < 1.01f);

    // VACUUM leaves every intensity exactly untouched, at every angle - bit-identical to applying
    // no correction at all.
    const auto vac = FlightPathAttenuation::Build(FlightPathMedium::Vacuum, D_mm, lambda_of(3.76));
    CHECK(vac.d_over_L == 0.0f);
    CHECK(vac.Factor(1.0f) == 1.0f);
    CHECK(vac.Factor(0.5f) == 1.0f);
    CHECK(vac.Factor(0.1f) == 1.0f);
    // Same for a file that states no distance or no wavelength.
    CHECK(FlightPathAttenuation::Build(FlightPathMedium::Air, 0.0, lambda_of(12.4)).d_over_L == 0.0f);
    CHECK(FlightPathAttenuation::Build(FlightPathMedium::Air, D_mm, 0.0).d_over_L == 0.0f);

    // What the report quotes as the worth of the assumption. On an untilted detector the correction
    // is a pure function of resolution, so its whole effect on merged data is a Wilson-B shift; the
    // sign is negative because lifting the high-angle data flattens the fall-off. Measured on real
    // data at three geometries, this predicts the observed shift to within 8%.
    const double dB_hard = air(13.0).WilsonBShift_A2(1.28, 50.0, lambda_of(13.0));
    const double dB_soft = air(3.76).WilsonBShift_A2(3.02, 50.0, lambda_of(3.76));
    CHECK(dB_hard < 0.0);
    CHECK(std::fabs(dB_hard) < 1.0);          // hard X-rays: nothing a user could read off the data
    CHECK(dB_soft < -5.0);                    // long wavelength: the dominant correction in the run
    CHECK(std::fabs(dB_soft) > std::fabs(dB_hard) * 10.0);
    // Vacuum is worth exactly nothing, which is what makes the report line meaningful.
    CHECK(vac.WilsonBShift_A2(3.02, 50.0, lambda_of(3.76)) == 0.0);
}

TEST_CASE("BraggPrediction_11keV") {
    DiffractionExperiment experiment(DetJF4M());
    experiment.DetectorDistance_mm(100.0).BeamX_pxl(1500.0).BeamY_pxl(1000.0)
    .IncidentEnergy_keV(11.0);

    DiffractionGeometry geom = experiment.GetDiffractionGeometry();

    CrystalLattice lattice(Coord{20, 10, 0}, Coord{-20, 40, 0},
                           Coord{0, 0, 200});

    BraggPredictionSettings settings{
        .high_res_A = 2.0,
        .ewald_dist_cutoff = 0.001,
        .max_h = 40, .max_k = 40, .max_l = 40
    };
    BraggPrediction prediction;

    int count = prediction.Calc(experiment, lattice, settings);
    REQUIRE(count > 0);

    for (int i = 0; i < count; i++) {
        auto r = prediction.GetReflections().at(i);
        auto recip = r.h * lattice.Astar() + r.k * lattice.Bstar() + r.l * lattice.Cstar();

        REQUIRE(std::abs(r.h) < settings.max_h );
        REQUIRE(std::abs(r.k) < settings.max_k );
        REQUIRE(std::abs(r.l) < settings.max_l );
        REQUIRE(r.d >= settings.high_res_A);
        REQUIRE(r.d == Catch::Approx(1/std::sqrt(recip * recip)).margin(0.01f));
        REQUIRE(r.dist_ewald == Catch::Approx(std::abs(geom.DistFromEwaldSphere(recip))).epsilon(1e-4));
        auto [x,y] = geom.RecipToDetector(recip);
        REQUIRE(r.predicted_x == Catch::Approx(x).margin(0.01));
        REQUIRE(r.predicted_y == Catch::Approx(y).margin(0.01));
    }
}

TEST_CASE("BraggPrediction_15keV") {
    DiffractionExperiment experiment(DetJF4M());
    experiment.DetectorDistance_mm(100.0).BeamX_pxl(1500.0).BeamY_pxl(1000.0)
    .IncidentEnergy_keV(15.0);

    DiffractionGeometry geom = experiment.GetDiffractionGeometry();

    CrystalLattice lattice(Coord{20, 10, 0}, Coord{-20, 40, 0},
                           Coord{0, 0, 200});

    BraggPredictionSettings settings{
        .high_res_A = 2.0,
        .ewald_dist_cutoff = 0.001,
        .max_h = 40, .max_k = 40, .max_l = 40
    };
    BraggPrediction prediction;

    int count = prediction.Calc(experiment, lattice, settings);
    REQUIRE(count > 0);

    for (int i = 0; i < count; i++) {
        auto r = prediction.GetReflections().at(i);
        auto recip = r.h * lattice.Astar() + r.k * lattice.Bstar() + r.l * lattice.Cstar();

        REQUIRE(std::abs(r.h) < settings.max_h );
        REQUIRE(std::abs(r.k) < settings.max_k );
        REQUIRE(std::abs(r.l) < settings.max_l );
        REQUIRE(r.d >= settings.high_res_A);
        REQUIRE(r.d == Catch::Approx(1/std::sqrt(recip * recip)).margin(0.01f));
        REQUIRE(r.dist_ewald == Catch::Approx(std::abs(geom.DistFromEwaldSphere(recip))).epsilon(1e-3));
        auto [x,y] = geom.RecipToDetector(recip);
        REQUIRE(r.predicted_x == Catch::Approx(x).margin(0.01));
        REQUIRE(r.predicted_y == Catch::Approx(y).margin(0.01));
    }
}

TEST_CASE("BraggPrediction_Rot1_Rot2") {
    DiffractionExperiment experiment(DetJF4M());
    experiment.DetectorDistance_mm(100.0).BeamX_pxl(1500.0).BeamY_pxl(1000.0)
    .PoniRot1_rad(2.0/180.0 * M_PI).PoniRot2_rad(3.0/180.0 * M_PI)
    .IncidentEnergy_keV(11.0);

    DiffractionGeometry geom = experiment.GetDiffractionGeometry();

    CrystalLattice lattice(Coord{20, 10, 0}, Coord{-20, 40, 0},
                           Coord{0, 0, 200});

    BraggPredictionSettings settings{
        .high_res_A = 2.0,
        .ewald_dist_cutoff = 0.001,
        .max_h = 40, .max_k = 40, .max_l = 40
    };
    BraggPrediction prediction;

    int count = prediction.Calc(experiment, lattice, settings);
    REQUIRE(count > 0);

    for (int i = 0; i < count; i++) {
        auto r = prediction.GetReflections().at(i);
        auto recip = r.h * lattice.Astar() + r.k * lattice.Bstar() + r.l * lattice.Cstar();

        REQUIRE(std::abs(r.h) < settings.max_h );
        REQUIRE(std::abs(r.k) < settings.max_k );
        REQUIRE(std::abs(r.l) < settings.max_l );
        REQUIRE(r.d >= settings.high_res_A);
        REQUIRE(r.d == Catch::Approx(1/std::sqrt(recip * recip)).margin(0.01f));
        REQUIRE(r.dist_ewald == Catch::Approx(std::abs(geom.DistFromEwaldSphere(recip))).epsilon(1e-4));
        auto [x,y] = geom.RecipToDetector(recip);
        REQUIRE(r.predicted_x == Catch::Approx(x).margin(0.01));
        REQUIRE(r.predicted_y == Catch::Approx(y).margin(0.01));
    }
}

TEST_CASE("BraggPrediction_backscattering") {
    DiffractionExperiment experiment(DetJF9M());
    experiment.DetectorDistance_mm(120.0).BeamX_pxl(1500.0).BeamY_pxl(1000.0)
        .IncidentEnergy_keV(3.0/WVL_1A_IN_KEV);

    // Orthogonal basis, sufficient to test parity rules
    CrystalLattice lattice(
        Coord{40, 0, 0},
        Coord{0, 50, 0},
        Coord{0, 0, 60}
    );

    BraggPredictionSettings settings{
        .high_res_A = 3.0f,
        .ewald_dist_cutoff = 0.1f, // Very large cutoff, to be able to see as many reflections as possible
        .max_h = 50, .max_k = 50, .max_l = 50
    };

    BraggPrediction pred;
    int count = pred.Calc(experiment, lattice, settings);
    REQUIRE(count > 0);
    for (const auto& r : pred.GetReflections()) {
        if (r.d == 0) break;
        REQUIRE(r.d > 3.0 / sqrt(2.0));
    }
}

TEST_CASE("BraggPrediction_systematic_absences") {
    DiffractionExperiment experiment(DetJF4M());
    experiment.DetectorDistance_mm(120.0).BeamX_pxl(1500.0).BeamY_pxl(1000.0)
        .IncidentEnergy_keV(12.0);

    // Orthogonal basis, sufficient to test parity rules
    CrystalLattice lattice(
        Coord{40, 0, 0},
        Coord{0, 50, 0},
        Coord{0, 0, 60}
    );

    BraggPredictionSettings settings{
        .high_res_A = 3.0f,
        .ewald_dist_cutoff = 0.1f, // Very large cutoff, to be able to see as many reflections as possible
        .max_h = 50, .max_k = 50, .max_l = 50
    };

    BraggPrediction pred;

    SECTION("I centering") {
        // 1) Body-centered I: reflections with h+k+l odd must be absent
        settings.centering = 'I';
        int count_I = pred.Calc(experiment, lattice, settings);
        REQUIRE(count_I > 0);
        for (const auto& r : pred.GetReflections()) {
            if (r.d == 0) break; // ignore unfilled tail if any
            REQUIRE(((r.h + r.k + r.l) % 2) == 0);
        }
    }

    SECTION ("F centering") {
        // 2) Face-centered F: h,k,l all even or all odd
        settings.centering = 'F';

        int count_F = pred.Calc(experiment, lattice, settings);
        REQUIRE(count_F > 0);
        for (const auto& r : pred.GetReflections()) {
            if (r.d == 0) break;
            const bool he = (r.h & 1) == 0, ke = (r.k & 1) == 0, le = (r.l & 1) == 0;
            const bool all_even = he && ke && le;
            const bool all_odd = (!he) && (!ke) && (!le);
            REQUIRE((all_even || all_odd));
        }
    }

    SECTION("R centering") {
        settings.centering = 'R';
        int count_R = pred.Calc(experiment, lattice, settings);
        REQUIRE(count_R > 0);

        // R (hexagonal setting): -h + k + l = 3n
        for (const auto& r : pred.GetReflections()) {
            if (r.d == 0) break;
            int cond = (-r.h + r.k + r.l) % 3;
            if (cond < 0) cond += 3;
            REQUIRE(cond == 0);
        }
    }
}

#ifdef JFJOCH_USE_CUDA
#include "../image_analysis/bragg_prediction/BraggPredictionGPU.h"
#include "../image_analysis/bragg_prediction/BraggPredictionRotGPU.h"
#include "../image_analysis/bragg_prediction/BraggPredictionRot.h"
#include <map>
#include <algorithm>

TEST_CASE("BraggPredictionGPU") {
    DiffractionExperiment experiment(DetJF4M());
    experiment.DetectorDistance_mm(100.0).BeamX_pxl(1500.0).BeamY_pxl(1000.0)
    .IncidentEnergy_keV(13.0);

    DiffractionGeometry geom = experiment.GetDiffractionGeometry();

    CrystalLattice lattice(Coord{20, 10, 0}, Coord{-20, 40, 0},
                           Coord{0, 0, 200});

    BraggPredictionSettings settings{
        .high_res_A = 2.0,
        .ewald_dist_cutoff = 0.001,
        .max_h = 40, .max_k = 40, .max_l = 40
    };
    BraggPredictionGPU prediction;

    int count = prediction.Calc(experiment, lattice, settings);
    REQUIRE(count > 0);

    for (int i = 0; i < count; i++) {
        auto r = prediction.GetReflections().at(i);
        auto recip = r.h * lattice.Astar() + r.k * lattice.Bstar() + r.l * lattice.Cstar();

        REQUIRE(std::abs(r.h) < settings.max_h );
        REQUIRE(std::abs(r.k) < settings.max_k );
        REQUIRE(std::abs(r.l) < settings.max_l );
        REQUIRE(r.d >= settings.high_res_A);
        REQUIRE(r.d == Catch::Approx(1/std::sqrt(recip * recip)).margin(0.01f));
        REQUIRE(r.dist_ewald == Catch::Approx(std::abs(geom.DistFromEwaldSphere(recip))).epsilon(2e-2));
        auto [x,y] = geom.RecipToDetector(recip);
        REQUIRE(r.predicted_x == Catch::Approx(x).margin(0.05));
        REQUIRE(r.predicted_y == Catch::Approx(y).margin(0.05));
    }
}

TEST_CASE("BraggPredictionGPU_Rot1_Rot2") {
    DiffractionExperiment experiment(DetJF4M());
    experiment.DetectorDistance_mm(100.0).BeamX_pxl(1500.0).BeamY_pxl(1000.0)
    .PoniRot1_rad(2.0/180.0 * M_PI).PoniRot2_rad(3.0/180.0 * M_PI)
    .IncidentEnergy_keV(11.0);

    DiffractionGeometry geom = experiment.GetDiffractionGeometry();

    CrystalLattice lattice(Coord{20, 10, 0}, Coord{-20, 40, 0},
                           Coord{0, 0, 200});

    BraggPredictionSettings settings{
        .high_res_A = 2.0,
        .ewald_dist_cutoff = 0.001,
        .max_h = 40, .max_k = 40, .max_l = 40
    };
    BraggPredictionGPU prediction;

    int count = prediction.Calc(experiment, lattice, settings);
    REQUIRE(count > 0);

    for (int i = 0; i < count; i++) {
        auto r = prediction.GetReflections().at(i);
        auto recip = r.h * lattice.Astar() + r.k * lattice.Bstar() + r.l * lattice.Cstar();

        REQUIRE(std::abs(r.h) < settings.max_h );
        REQUIRE(std::abs(r.k) < settings.max_k );
        REQUIRE(std::abs(r.l) < settings.max_l );
        REQUIRE(r.d >= settings.high_res_A);
        REQUIRE(r.d == Catch::Approx(1/std::sqrt(recip * recip)).margin(0.01f));
        REQUIRE(r.dist_ewald == Catch::Approx(std::abs(geom.DistFromEwaldSphere(recip))).epsilon(1e-3));
        auto [x,y] = geom.RecipToDetector(recip);
        REQUIRE(r.predicted_x == Catch::Approx(x).margin(0.05));
        REQUIRE(r.predicted_y == Catch::Approx(y).margin(0.05));
    }
}

TEST_CASE("BraggPredictionGPU_systematic_absences") {
    DiffractionExperiment experiment(DetJF4M());
    experiment.DetectorDistance_mm(120.0).BeamX_pxl(1500.0).BeamY_pxl(1000.0)
        .IncidentEnergy_keV(12.0);

    CrystalLattice lattice(
        Coord{40, 0, 0},
        Coord{0, 50, 0},
        Coord{0, 0, 60}
    );

    BraggPredictionSettings settings{
        .high_res_A = 3.0f,
        .ewald_dist_cutoff = 0.1f,
        .max_h = 50, .max_k = 50, .max_l = 50
    };

    BraggPredictionGPU pred;
    SECTION ("I centering") {
        // 1) Body-centered I
        settings.centering = 'I';

        int count_I = pred.Calc(experiment, lattice, settings);
        REQUIRE(count_I > 0);
        for (const auto& r : pred.GetReflections()) {
            if (r.d == 0) break;
            REQUIRE(((r.h + r.k + r.l) % 2) == 0);
        }
    }

    SECTION ("F centering") {
        // 2) Face-centered F
        settings.centering = 'F';
        int count_F = pred.Calc(experiment, lattice, settings);
        REQUIRE(count_F > 0);
        for (const auto& r : pred.GetReflections()) {
            if (r.d == 0) break;
            const bool he = (r.h & 1) == 0, ke = (r.k & 1) == 0, le = (r.l & 1) == 0;
            const bool all_even = he && ke && le;
            const bool all_odd = (!he) && (!ke) && (!le);
            REQUIRE((all_even || all_odd));
        }
    }

    SECTION("R centering") {
        settings.centering = 'R';
        int count_R = pred.Calc(experiment, lattice, settings);
        REQUIRE(count_R > 0);

        // R (hexagonal setting): -h + k + l = 3n
        for (const auto& r : pred.GetReflections()) {
            if (r.d == 0) break;
            int cond = (-r.h + r.k + r.l) % 3;
            if (cond < 0) cond += 3;
            REQUIRE(cond == 0);
        }
    }
}

TEST_CASE("BraggPredictionGPU_backscattering") {
    DiffractionExperiment experiment(DetJF9M());
    experiment.DetectorDistance_mm(120.0).BeamX_pxl(1500.0).BeamY_pxl(1000.0)
        .IncidentEnergy_keV(3.0/WVL_1A_IN_KEV);

    // Orthogonal basis, sufficient to test parity rules
    CrystalLattice lattice(
        Coord{40, 0, 0},
        Coord{0, 50, 0},
        Coord{0, 0, 60}
    );

    BraggPredictionSettings settings{
        .high_res_A = 3.0f,
        .ewald_dist_cutoff = 0.1f, // Very large cutoff, to be able to see as many reflections as possible
        .max_h = 50, .max_k = 50, .max_l = 50
    };

    BraggPredictionGPU pred;
    int count = pred.Calc(experiment, lattice, settings);
    REQUIRE(count > 0);
    for (const auto& r : pred.GetReflections()) {
        if (r.d == 0) break;
        REQUIRE(r.d > 3.0 / sqrt(2.0));
    }
}

TEST_CASE("BraggPrediction_CPU_GPU_consistency_tilted") {
    // Verify CPU and GPU implementations produce identical results with tilted detector
    DiffractionExperiment experiment(DetJF4M());
    experiment.DetectorDistance_mm(100.0).BeamX_pxl(1500.0).BeamY_pxl(1000.0)
        .PoniRot1_rad(0.04).PoniRot2_rad(-0.025)
        .IncidentEnergy_keV(12.0);

    CrystalLattice lattice(Coord{30, 10, 0}, Coord{-15, 45, 0}, Coord{0, 0, 150});

    BraggPredictionSettings settings{
        .high_res_A = 2.0,
        .ewald_dist_cutoff = 0.0015,
        .max_h = 30, .max_k = 30, .max_l = 30
    };

    BraggPrediction cpu_pred;
    BraggPredictionGPU gpu_pred;

    int cpu_count = cpu_pred.Calc(experiment, lattice, settings);
    int gpu_count = gpu_pred.Calc(experiment, lattice, settings);

    REQUIRE(cpu_count > 0);
    REQUIRE(gpu_count > 0);

    // Build map of GPU reflections by hkl
    std::map<std::tuple<int,int,int>, const Reflection*> gpu_refl_map;
    for (int i = 0; i < gpu_count; ++i) {
        const auto& r = gpu_pred.GetReflections().at(i);
        gpu_refl_map[{r.h, r.k, r.l}] = &r;
    }

    // Check that each CPU reflection has a matching GPU reflection
    int matched = 0;
    float min_corr = 1.0f;
    float max_flight = 1.0f;
    for (int i = 0; i < cpu_count; ++i) {
        const auto& cpu_r = cpu_pred.GetReflections().at(i);
        auto key = std::make_tuple(cpu_r.h, cpu_r.k, cpu_r.l);
        auto it = gpu_refl_map.find(key);
        if (it != gpu_refl_map.end()) {
            const auto& gpu_r = *it->second;
            CHECK(cpu_r.predicted_x == Catch::Approx(gpu_r.predicted_x).margin(0.1));
            CHECK(cpu_r.predicted_y == Catch::Approx(gpu_r.predicted_y).margin(0.1));
            CHECK(cpu_r.d == Catch::Approx(gpu_r.d).margin(0.01));
            // Both halves of the correction are part of the prediction, not a downstream product:
            // the sensor-efficiency term lived on the CPU path alone for a while because nothing here
            // compared it, and it is now a field of its own - so it is compared as a field of its own.
            CHECK(cpu_r.prescaling_corr == Catch::Approx(gpu_r.prescaling_corr).epsilon(1e-4));
            CHECK(cpu_r.qe_corr == Catch::Approx(gpu_r.qe_corr).epsilon(1e-4));
            CHECK(cpu_r.flight_corr == Catch::Approx(gpu_r.flight_corr).epsilon(1e-4));
            CHECK(cpu_r.image_scale_corr == Catch::Approx(gpu_r.image_scale_corr).epsilon(1e-4));
            min_corr = std::min(min_corr, cpu_r.qe_corr);
            max_flight = std::max(max_flight, cpu_r.flight_corr);
            matched++;
        }
    }

    // Most reflections should match (allow for some numerical differences at boundaries)
    CHECK(matched > cpu_count * 0.95);
    // ... and the comparison above must not be vacuous: on this geometry (320 um Si at 12 keV,
    // reflections out to 2 A) the sensor correction is several per cent, so a qe_corr that stayed
    // at 1 on both sides would mean the correction had been dropped from BOTH paths. It is checked
    // on qe_corr and not on prescaling_corr, which no longer carries it.
    CHECK(min_corr < 0.99f);
    // The same for the air term, which runs the other way: at 12 keV over this distance the air
    // correction reaches a few tenths of a per cent at the detector corner, so an flight_corr pinned at
    // 1 on both sides would mean it had been dropped from BOTH paths rather than agreeing.
    CHECK(max_flight > 1.0f);
}

TEST_CASE("BraggPredictionRot_CPU_GPU_consistency_tilted") {
    // The rotation counterpart of the test above. Same purpose: every per-reflection quantity the
    // two implementations both produce has to agree - the Lorentz-polarization factor and the sensor
    // efficiency each in their own field, so neither can hide behind the other in a product.
    DiffractionExperiment experiment(DetJF4M());
    experiment.DetectorDistance_mm(100.0).BeamX_pxl(1500.0).BeamY_pxl(1000.0)
        .PoniRot1_rad(0.04).PoniRot2_rad(-0.025)
        .IncidentEnergy_keV(12.0)
        .Goniometer(GoniometerAxis("omega", 0.0f, 0.1f, Coord(-1, 0, 0), {}));

    CrystalLattice lattice(Coord{30, 10, 0}, Coord{-15, 45, 0}, Coord{0, 0, 150});

    BraggPredictionSettings settings{
        .high_res_A = 2.0,
        .ewald_dist_cutoff = 0.0015,
        .max_h = 30, .max_k = 30, .max_l = 30
    };

    BraggPredictionRot cpu_pred;
    BraggPredictionRotGPU gpu_pred;

    int cpu_count = cpu_pred.Calc(experiment, lattice, settings);
    int gpu_count = gpu_pred.Calc(experiment, lattice, settings);

    REQUIRE(cpu_count > 0);
    REQUIRE(gpu_count > 0);

    std::map<std::tuple<int,int,int>, const Reflection*> gpu_refl_map;
    for (int i = 0; i < gpu_count; ++i) {
        const auto& r = gpu_pred.GetReflections().at(i);
        gpu_refl_map[{r.h, r.k, r.l}] = &r;
    }

    int matched = 0;
    float min_corr = 1.0f;
    float max_flight = 1.0f;
    for (int i = 0; i < cpu_count; ++i) {
        const auto& cpu_r = cpu_pred.GetReflections().at(i);
        auto it = gpu_refl_map.find(std::make_tuple(cpu_r.h, cpu_r.k, cpu_r.l));
        if (it != gpu_refl_map.end()) {
            const auto& gpu_r = *it->second;
            CHECK(cpu_r.predicted_x == Catch::Approx(gpu_r.predicted_x).margin(0.1));
            CHECK(cpu_r.predicted_y == Catch::Approx(gpu_r.predicted_y).margin(0.1));
            CHECK(cpu_r.d == Catch::Approx(gpu_r.d).margin(0.01));
            CHECK(cpu_r.prescaling_corr == Catch::Approx(gpu_r.prescaling_corr).epsilon(1e-3));
            CHECK(cpu_r.qe_corr == Catch::Approx(gpu_r.qe_corr).epsilon(1e-3));
            CHECK(cpu_r.flight_corr == Catch::Approx(gpu_r.flight_corr).epsilon(1e-3));
            min_corr = std::min(min_corr, cpu_r.qe_corr);
            max_flight = std::max(max_flight, cpu_r.flight_corr);
            matched++;
        }
    }

    CHECK(matched > cpu_count * 0.95);
    CHECK(min_corr < 0.99f);
    CHECK(max_flight > 1.0f);
}
#endif