/*
 * Copyright (C) 2022 NHR@FAU, University Erlangen-Nuremberg.
 * All rights reserved. This file is part of nusif-solver.
 * Use of this source code is governed by a MIT style
 * license that can be found in the LICENSE file.
 */
#include <float.h>
#include <math.h>
#include <mpi.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>

#include "allocate.h"
#include "parameter.h"
#include "solver.h"
#include "util.h"

#define P(i, j)   p[(j) * (imax + 2) + (i)]
#define F(i, j)   f[(j) * (imax + 2) + (i)]
#define G(i, j)   g[(j) * (imax + 2) + (i)]
#define U(i, j)   u[(j) * (imax + 2) + (i)]
#define V(i, j)   v[(j) * (imax + 2) + (i)]
#define RHS(i, j) rhs[(j) * (imax + 2) + (i)]

static int sizeOfRank(int rank, int size, int N)
{
    return N / size + ((N % size > rank) ? 1 : 0);
}

static void print(Solver* solver, double* grid)
{
    int imax = solver->imax;

    for (int i = 0; i < solver->size; i++) {
        if (i == solver->rank) {
            printf("### RANK %d "
                   "#######################################################\n",
                solver->rank);
            for (int j = 0; j < solver->jmaxLocal + 2; j++) {
                printf("%02d: ", j);
                for (int i = 0; i < solver->imax + 2; i++) {
                    printf("%12.8f  ", grid[j * (imax + 2) + i]);
                }
                printf("\n");
            }
            fflush(stdout);
        }
        MPI_Barrier(MPI_COMM_WORLD);
    }
}

static void exchange(Solver* solver, double* grid)
{
    MPI_Request requests[4] = { MPI_REQUEST_NULL,
        MPI_REQUEST_NULL,
        MPI_REQUEST_NULL,
        MPI_REQUEST_NULL };

    /* exchange ghost cells with top neighbor */
    if (solver->rank + 1 < solver->size) {
        int top     = solver->rank + 1;
        double* src = grid + (solver->jmaxLocal) * (solver->imax + 2) + 1;
        double* dst = grid + (solver->jmaxLocal + 1) * (solver->imax + 2) + 1;

        MPI_Isend(src, solver->imax, MPI_DOUBLE, top, 1, MPI_COMM_WORLD, &requests[0]);
        MPI_Irecv(dst, solver->imax, MPI_DOUBLE, top, 2, MPI_COMM_WORLD, &requests[1]);
    }

    /* exchange ghost cells with bottom neighbor */
    if (solver->rank > 0) {
        int bottom  = solver->rank - 1;
        double* src = grid + (solver->imax + 2) + 1;
        double* dst = grid + 1;

        MPI_Isend(src, solver->imax, MPI_DOUBLE, bottom, 2, MPI_COMM_WORLD, &requests[2]);
        MPI_Irecv(dst, solver->imax, MPI_DOUBLE, bottom, 1, MPI_COMM_WORLD, &requests[3]);
    }

    MPI_Waitall(4, requests, MPI_STATUSES_IGNORE);
}

static void shift(Solver* solver)
{
    MPI_Request requests[2] = { MPI_REQUEST_NULL, MPI_REQUEST_NULL };
    double* g               = solver->g;

    /* shift G */
    /* receive ghost cells from bottom neighbor */
    if (solver->rank > 0) {
        int bottom = solver->rank - 1;
        MPI_Irecv(g + 1,
            solver->imax,
            MPI_DOUBLE,
            bottom,
            0,
            MPI_COMM_WORLD,
            &requests[0]);
    }

    if (solver->rank + 1 < solver->size) {
        int top     = solver->rank + 1;
        double* buf = g + (solver->jmaxLocal) * (solver->imax + 2) + 1;
        /* send ghost cells to top neighbor */
        MPI_Isend(buf, solver->imax, MPI_DOUBLE, top, 0, MPI_COMM_WORLD, &requests[1]);
    }

    MPI_Waitall(2, requests, MPI_STATUSES_IGNORE);
}

static void gatherArray(
    Solver* solver, int cnt, int* rcvCounts, int* displs, double* src, double* dst)
{
    double* sendbuffer = src + (solver->imax + 2);

    if (solver->rank == 0) {
        sendbuffer = src;
    }

    MPI_Gatherv(sendbuffer,
        cnt,
        MPI_DOUBLE,
        dst,
        rcvCounts,
        displs,
        MPI_DOUBLE,
        0,
        MPI_COMM_WORLD);
}

void collectResult(Solver* solver)
{
    double* p = NULL;
    double* u = NULL;
    double* v = NULL;
    int *rcvCounts, *displs;
    int cnt = solver->jmaxLocal * (solver->imax + 2);

    if (solver->rank == 0) {
        p = allocate(64, (solver->imax + 2) * (solver->jmax + 2) * sizeof(double));
        u = allocate(64, (solver->imax + 2) * (solver->jmax + 2) * sizeof(double));
        v = allocate(64, (solver->imax + 2) * (solver->jmax + 2) * sizeof(double));
        rcvCounts    = (int*)malloc(solver->size * sizeof(int));
        displs       = (int*)malloc(solver->size * sizeof(int));
    }

    if (solver->rank == 0 && solver->size == 1) {
        cnt = (solver->jmaxLocal + 2) * (solver->imax + 2);
    } else if (solver->rank == 0 || solver->rank == (solver->size - 1)) {
        cnt = (solver->jmaxLocal + 1) * (solver->imax + 2);
    }

    MPI_Gather(&cnt, 1, MPI_INTEGER, rcvCounts, 1, MPI_INTEGER, 0, MPI_COMM_WORLD);

    if (solver->rank == 0) {
        displs[0]    = 0;
        int cursor   = rcvCounts[0];

        for (int i = 1; i < solver->size; i++) {
            displs[i]    = cursor;
            cursor += rcvCounts[i];
        }
    }

    gatherArray(solver, cnt, rcvCounts, displs, solver->p, p);
    gatherArray(solver, cnt, rcvCounts, displs, solver->u, u);
    gatherArray(solver, cnt, rcvCounts, displs, solver->v, v);

    if (solver->rank == 0) {
        writeResult(solver, p, u, v);
        free(p);
        free(u);
        free(v);
    }
}

static void printConfig(Solver* solver)
{
    if (solver->rank == 0) {
        printf("Parameters for #%s#\n", solver->problem);
        printf("Boundary conditions Left:%d Right:%d Bottom:%d Top:%d\n",
            solver->bcLeft,
            solver->bcRight,
            solver->bcBottom,
            solver->bcTop);
        printf("\tReynolds number: %.2f\n", solver->re);
        printf("\tGx Gy: %.2f %.2f\n", solver->gx, solver->gy);
        printf("Geometry data:\n");
        printf("\tDomain box size (x, y): %.2f, %.2f\n",
            solver->xlength,
            solver->ylength);
        printf("\tCells (x, y): %d, %d\n", solver->imax, solver->jmax);
        printf("Timestep parameters:\n");
        printf("\tDefault stepsize: %.2f, Final time %.2f\n", solver->dt, solver->te);
        printf("\tdt bound: %.6f\n", solver->dtBound);
        printf("\tTau factor: %.2f\n", solver->tau);
        printf("Iterative solver parameters:\n");
        printf("\tMax iterations: %d\n", solver->itermax);
        printf("\tepsilon (stopping tolerance) : %f\n", solver->eps);
        printf("\tgamma factor: %f\n", solver->gamma);
        printf("\tomega (SOR relaxation): %f\n", solver->omega);
        printf("Communication parameters:\n");
    }
    for (int i = 0; i < solver->size; i++) {
        if (i == solver->rank) {
            printf("\tRank %d of %d\n", solver->rank, solver->size);
            printf("\tLocal domain size: %dx%d\n", solver->imax, solver->jmaxLocal);
            fflush(stdout);
        }
    }
}

void initSolver(Solver* solver, Parameter* params)
{
    MPI_Comm_rank(MPI_COMM_WORLD, &(solver->rank));
    MPI_Comm_size(MPI_COMM_WORLD, &(solver->size));
    solver->problem   = params->name;
    solver->bcLeft    = params->bcLeft;
    solver->bcRight   = params->bcRight;
    solver->bcBottom  = params->bcBottom;
    solver->bcTop     = params->bcTop;
    solver->imax      = params->imax;
    solver->jmax      = params->jmax;
    solver->jmaxLocal = sizeOfRank(solver->rank, solver->size, solver->jmax);
    solver->xlength   = params->xlength;
    solver->ylength   = params->ylength;
    solver->dx        = params->xlength / params->imax;
    solver->dy        = params->ylength / params->jmax;
    solver->eps       = params->eps;
    solver->omega     = params->omg;
    solver->itermax   = params->itermax;
    solver->re        = params->re;
    solver->gx        = params->gx;
    solver->gy        = params->gy;
    solver->dt        = params->dt;
    solver->te        = params->te;
    solver->tau       = params->tau;
    solver->gamma     = params->gamma;

    int imax        = solver->imax;
    int jmaxLocal   = solver->jmaxLocal;
    size_t bytesize = (imax + 2) * (jmaxLocal + 2) * sizeof(double);
    solver->u       = allocate(64, bytesize);
    solver->v       = allocate(64, bytesize);
    solver->p       = allocate(64, bytesize);
    solver->rhs     = allocate(64, bytesize);
    solver->f       = allocate(64, bytesize);
    solver->g       = allocate(64, bytesize);

    for (int i = 0; i < (imax + 2) * (jmaxLocal + 2); i++) {
        solver->u[i]   = params->u_init;
        solver->v[i]   = params->v_init;
        solver->p[i]   = params->p_init;
        solver->rhs[i] = 0.0;
        solver->f[i]   = 0.0;
        solver->g[i]   = 0.0;
    }

    double dx          = solver->dx;
    double dy          = solver->dy;
    double invSquareSum = 1.0 / (dx * dx) + 1.0 / (dy * dy);
    solver->dtBound     = 0.5 * solver->re * 1.0 / invSquareSum;
#ifdef VERBOSE
    printConfig(solver);
#endif
}

void computeRHS(Solver* solver)
{
    int imax      = solver->imax;
    int jmaxLocal = solver->jmaxLocal;
    double idx    = 1.0 / solver->dx;
    double idy    = 1.0 / solver->dy;
    double idt    = 1.0 / solver->dt;
    double* rhs   = solver->rhs;
    double* f     = solver->f;
    double* g     = solver->g;

    shift(solver);

    for (int j = 1; j < jmaxLocal + 1; j++) {
        for (int i = 1; i < imax + 1; i++) {
            RHS(i, j) = ((F(i, j) - F(i - 1, j)) * idx + (G(i, j) - G(i, j - 1)) * idy) *
                        idt;
        }
    }
}

void solve(Solver* solver)
{
    int imax      = solver->imax;
    int jmax      = solver->jmax;
    int jmaxLocal = solver->jmaxLocal;
    double eps    = solver->eps;
    int itermax   = solver->itermax;
    double dx2    = solver->dx * solver->dx;
    double dy2    = solver->dy * solver->dy;
    double idx2   = 1.0 / dx2;
    double idy2   = 1.0 / dy2;
    double factor = solver->omega * 0.5 * (dx2 * dy2) / (dx2 + dy2);
    double* p     = solver->p;
    double* rhs   = solver->rhs;
    double epssq  = eps * eps;
    int it        = 0;
    double res    = 1.0;

    while ((res >= epssq) && (it < itermax)) {
        res = 0.0;
        exchange(solver, p);

        for (int j = 1; j < jmaxLocal + 1; j++) {
            for (int i = 1; i < imax + 1; i++) {

                double r = RHS(i, j) -
                           ((P(i + 1, j) - 2.0 * P(i, j) + P(i - 1, j)) * idx2 +
                               (P(i, j + 1) - 2.0 * P(i, j) + P(i, j - 1)) * idy2);

                P(i, j) -= (factor * r);
                res += (r * r);
            }
        }

        if (solver->rank == 0) {
            for (int i = 1; i < imax + 1; i++) {
                P(i, 0) = P(i, 1);
            }
        }

        if (solver->rank == (solver->size - 1)) {
            for (int i = 1; i < imax + 1; i++) {
                P(i, jmaxLocal + 1) = P(i, jmaxLocal);
            }
        }

        for (int j = 1; j < jmaxLocal + 1; j++) {
            P(0, j)        = P(1, j);
            P(imax + 1, j) = P(imax, j);
        }

        MPI_Allreduce(MPI_IN_PLACE, &res, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
        res = res / (double)(imax * jmax);
#ifdef DEBUG
        if (solver->rank == 0) {
            printf("%d Residuum: %e\n", it, res);
        }
#endif
        it++;
    }

#ifdef VERBOSE
    if (solver->rank == 0) {
        printf("Solver took %d iterations to reach %f\n", it, sqrt(res));
    }
#endif
}

static double maxElement(Solver* solver, double* m)
{
    int size      = (solver->imax + 2) * (solver->jmaxLocal + 2);
    double maxval = DBL_MIN;

    for (int i = 0; i < size; i++) {
        maxval = MAX(maxval, fabs(m[i]));
    }

    MPI_Allreduce(MPI_IN_PLACE, &maxval, 1, MPI_DOUBLE, MPI_MAX, MPI_COMM_WORLD);
    return maxval;
}

void normalizePressure(Solver* solver)
{
    int size    = (solver->imax + 2) * (solver->jmaxLocal + 2);
    double* p   = solver->p;
    double avgP = 0.0;

    for (int i = 0; i < size; i++) {
        avgP += p[i];
    }
    MPI_Allreduce(MPI_IN_PLACE, &avgP, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
    avgP /= (solver->imax + 2) * (solver->jmax + 2);

    for (int i = 0; i < size; i++) {
        p[i] = p[i] - avgP;
    }
}

void computeTimestep(Solver* solver)
{
    double dt   = solver->dtBound;
    double dx   = solver->dx;
    double dy   = solver->dy;
    double umax = maxElement(solver, solver->u);
    double vmax = maxElement(solver, solver->v);

    if (umax > 0) {
        dt = (dt > dx / umax) ? dx / umax : dt;
    }
    if (vmax > 0) {
        dt = (dt > dy / vmax) ? dy / vmax : dt;
    }

    solver->dt = dt * solver->tau;
}

void setBoundaryConditions(Solver* solver)
{
    int imax      = solver->imax;
    int jmaxLocal = solver->jmaxLocal;
    double* u     = solver->u;
    double* v     = solver->v;

    // Left boundary
    switch (solver->bcLeft) {
    case NOSLIP:
        for (int j = 1; j < jmaxLocal + 1; j++) {
            U(0, j) = 0.0;
            V(0, j) = -V(1, j);
        }
        break;
    case SLIP:
        for (int j = 1; j < jmaxLocal + 1; j++) {
            U(0, j) = 0.0;
            V(0, j) = V(1, j);
        }
        break;
    case OUTFLOW:
        for (int j = 1; j < jmaxLocal + 1; j++) {
            U(0, j) = U(1, j);
            V(0, j) = V(1, j);
        }
        break;
    case PERIODIC:
        break;
    }

    // Right boundary
    switch (solver->bcRight) {
    case NOSLIP:
        for (int j = 1; j < jmaxLocal + 1; j++) {
            U(imax, j)     = 0.0;
            V(imax + 1, j) = -V(imax, j);
        }
        break;
    case SLIP:
        for (int j = 1; j < jmaxLocal + 1; j++) {
            U(imax, j)     = 0.0;
            V(imax + 1, j) = V(imax, j);
        }
        break;
    case OUTFLOW:
        for (int j = 1; j < jmaxLocal + 1; j++) {
            U(imax, j)     = U(imax - 1, j);
            V(imax + 1, j) = V(imax, j);
        }
        break;
    case PERIODIC:
        break;
    }

    // Bottom boundary
    if (solver->rank == 0) {
        switch (solver->bcBottom) {
        case NOSLIP:
            for (int i = 1; i < imax + 1; i++) {
                V(i, 0) = 0.0;
                U(i, 0) = -U(i, 1);
            }
            break;
        case SLIP:
            for (int i = 1; i < imax + 1; i++) {
                V(i, 0) = 0.0;
                U(i, 0) = U(i, 1);
            }
            break;
        case OUTFLOW:
            for (int i = 1; i < imax + 1; i++) {
                U(i, 0) = U(i, 1);
                V(i, 0) = V(i, 1);
            }
            break;
        case PERIODIC:
            break;
        }
    }

    // Top boundary
    if (solver->rank == (solver->size - 1)) {
        switch (solver->bcTop) {
        case NOSLIP:
            for (int i = 1; i < imax + 1; i++) {
                V(i, jmaxLocal)     = 0.0;
                U(i, jmaxLocal + 1) = -U(i, jmaxLocal);
            }
            break;
        case SLIP:
            for (int i = 1; i < imax + 1; i++) {
                V(i, jmaxLocal)     = 0.0;
                U(i, jmaxLocal + 1) = U(i, jmaxLocal);
            }
            break;
        case OUTFLOW:
            for (int i = 1; i < imax + 1; i++) {
                U(i, jmaxLocal + 1) = U(i, jmaxLocal);
                V(i, jmaxLocal)     = V(i, jmaxLocal - 1);
            }
            break;
        case PERIODIC:
            break;
        }
    }
}

void setSpecialBoundaryCondition(Solver* solver)
{
    int imax      = solver->imax;
    int jmaxLocal = solver->jmaxLocal;
    double* u     = solver->u;

    if (strcmp(solver->problem, "dcavity") == 0) {
        if (solver->rank == (solver->size - 1)) {
            for (int i = 1; i < imax; i++) {
                U(i, jmaxLocal + 1) = 2.0 - U(i, jmaxLocal);
            }
        }
    } else if (strcmp(solver->problem, "canal") == 0) {
        double ylength = solver->ylength;
        double dy      = solver->dy;
        int rest       = solver->jmax % solver->size;
        int yc = solver->rank * (solver->jmax / solver->size) + MIN(rest, solver->rank);
        double ys = dy * (yc + 0.5);
        double y;

        /* printf("RANK %d yc: %d ys: %f\n", solver->rank, yc, ys); */

        for (int j = 1; j < jmaxLocal + 1; j++) {
            y       = ys + dy * (j - 0.5);
            U(0, j) = y * (ylength - y) * 4.0 / (ylength * ylength);
        }
    }
    /* print(solver, solver->u); */
}

void computeFG(Solver* solver)
{
    double* u        = solver->u;
    double* v        = solver->v;
    double* f        = solver->f;
    double* g        = solver->g;
    int imax         = solver->imax;
    int jmaxLocal    = solver->jmaxLocal;
    double gx        = solver->gx;
    double gy        = solver->gy;
    double gamma     = solver->gamma;
    double dt        = solver->dt;
    double inverseRe = 1.0 / solver->re;
    double inverseDx = 1.0 / solver->dx;
    double inverseDy = 1.0 / solver->dy;
    double du2dx, dv2dy, duvdx, duvdy;
    double du2dx2, du2dy2, dv2dx2, dv2dy2;

    exchange(solver, u);
    exchange(solver, v);

    for (int j = 1; j < jmaxLocal + 1; j++) {
        for (int i = 1; i < imax + 1; i++) {
            du2dx = inverseDx * 0.25 *
                        ((U(i, j) + U(i + 1, j)) * (U(i, j) + U(i + 1, j)) -
                            (U(i, j) + U(i - 1, j)) * (U(i, j) + U(i - 1, j))) +
                    gamma * inverseDx * 0.25 *
                        (fabs(U(i, j) + U(i + 1, j)) * (U(i, j) - U(i + 1, j)) +
                            fabs(U(i, j) + U(i - 1, j)) * (U(i, j) - U(i - 1, j)));

            duvdy = inverseDy * 0.25 *
                        ((V(i, j) + V(i + 1, j)) * (U(i, j) + U(i, j + 1)) -
                            (V(i, j - 1) + V(i + 1, j - 1)) * (U(i, j) + U(i, j - 1))) +
                    gamma * inverseDy * 0.25 *
                        (fabs(V(i, j) + V(i + 1, j)) * (U(i, j) - U(i, j + 1)) +
                            fabs(V(i, j - 1) + V(i + 1, j - 1)) *
                                (U(i, j) - U(i, j - 1)));

            du2dx2  = inverseDx * inverseDx * (U(i + 1, j) - 2.0 * U(i, j) + U(i - 1, j));
            du2dy2  = inverseDy * inverseDy * (U(i, j + 1) - 2.0 * U(i, j) + U(i, j - 1));
            F(i, j) = U(i, j) + dt * (inverseRe * (du2dx2 + du2dy2) - du2dx - duvdy + gx);

            duvdx = inverseDx * 0.25 *
                        ((U(i, j) + U(i, j + 1)) * (V(i, j) + V(i + 1, j)) -
                            (U(i - 1, j) + U(i - 1, j + 1)) * (V(i, j) + V(i - 1, j))) +
                    gamma * inverseDx * 0.25 *
                        (fabs(U(i, j) + U(i, j + 1)) * (V(i, j) - V(i + 1, j)) +
                            fabs(U(i - 1, j) + U(i - 1, j + 1)) *
                                (V(i, j) - V(i - 1, j)));

            dv2dy = inverseDy * 0.25 *
                        ((V(i, j) + V(i, j + 1)) * (V(i, j) + V(i, j + 1)) -
                            (V(i, j) + V(i, j - 1)) * (V(i, j) + V(i, j - 1))) +
                    gamma * inverseDy * 0.25 *
                        (fabs(V(i, j) + V(i, j + 1)) * (V(i, j) - V(i, j + 1)) +
                            fabs(V(i, j) + V(i, j - 1)) * (V(i, j) - V(i, j - 1)));

            dv2dx2  = inverseDx * inverseDx * (V(i + 1, j) - 2.0 * V(i, j) + V(i - 1, j));
            dv2dy2  = inverseDy * inverseDy * (V(i, j + 1) - 2.0 * V(i, j) + V(i, j - 1));
            G(i, j) = V(i, j) + dt * (inverseRe * (dv2dx2 + dv2dy2) - duvdx - dv2dy + gy);
        }
    }

    /* ----------------------------- boundary of F ---------------------------
     */
    for (int j = 1; j < jmaxLocal + 1; j++) {
        F(0, j)    = U(0, j);
        F(imax, j) = U(imax, j);
    }

    /* ----------------------------- boundary of G ---------------------------
     */
    if (solver->rank == 0) {
        for (int i = 1; i < imax + 1; i++) {
            G(i, 0) = V(i, 0);
        }
    }

    if (solver->rank == (solver->size - 1)) {
        for (int i = 1; i < imax + 1; i++) {
            G(i, jmaxLocal) = V(i, jmaxLocal);
        }
    }
}

void adaptUV(Solver* solver)
{
    int imax       = solver->imax;
    int jmaxLocal  = solver->jmaxLocal;
    double* p      = solver->p;
    double* u      = solver->u;
    double* v      = solver->v;
    double* f      = solver->f;
    double* g      = solver->g;
    double factorX = solver->dt / solver->dx;
    double factorY = solver->dt / solver->dy;

    for (int j = 1; j < jmaxLocal + 1; j++) {
        for (int i = 1; i < imax + 1; i++) {
            U(i, j) = F(i, j) - (P(i + 1, j) - P(i, j)) * factorX;
            V(i, j) = G(i, j) - (P(i, j + 1) - P(i, j)) * factorY;
        }
    }
}

void writeResult(Solver* solver, double* p, double* u, double* v)
{
    int imax  = solver->imax;
    int jmax  = solver->jmax;
    double dx = solver->dx;
    double dy = solver->dy;
    double x = 0.0, y = 0.0;

    FILE* fp;
    fp = fopen("pressure.dat", "w");

    if (fp == NULL) {
        printf("Error!\n");
        exit(EXIT_FAILURE);
    }

    for (int j = 1; j < jmax + 1; j++) {
        y = (double)(j - 0.5) * dy;
        for (int i = 1; i < imax + 1; i++) {
            x = (double)(i - 0.5) * dx;
            fprintf(fp, "%.2f %.2f %f\n", x, y, P(i, j));
        }
        fprintf(fp, "\n");
    }

    fclose(fp);

    fp = fopen("velocity.dat", "w");

    if (fp == NULL) {
        printf("Error!\n");
        exit(EXIT_FAILURE);
    }

    for (int j = 1; j < jmax + 1; j++) {
        y = dy * (j - 0.5);
        for (int i = 1; i < imax + 1; i++) {
            x            = dx * (i - 0.5);
            double vu  = (U(i, j) + U(i - 1, j)) / 2.0;
            double vv  = (V(i, j) + V(i, j - 1)) / 2.0;
            double len = sqrt((vu * vu) + (vv * vv));
            fprintf(fp, "%.2f %.2f %f %f %f\n", x, y, vu, vv, len);
        }
    }

    fclose(fp);
}