#include <cstdio>
#include <cstdlib>

#include "Vline_buffer.h"
#include "verilated_vcd_c.h"

using namespace std;

Vline_buffer* top;

vluint64_t main_time = 0;

double sc_time_stamp() {
    return main_time;
}

const int IMAGE_SIZE = 5;  // Number of columns in the test image
const int ROWS = 5;        // Number of rows in the test image
const int NUM_LINES = 3;   // Number of lines in the sliding window
const int DATA_WIDTH = 8;  // Pixel width in bits
const int PADDING = 1;     // Enable padding

int main(int argc, char** argv) {
    Verilated::commandArgs(argc, argv);

    top = new Vline_buffer;

    // Enable waveform tracing
    Verilated::traceEverOn(true);
    VerilatedVcdC* tfp = new VerilatedVcdC;
    top->trace(tfp, 99);
    tfp->open("line_buffer.vcd");

    // Initialize signals
    top->clk = 0;
    top->rst = 1;
    top->pixel_in = 0;
    top->pixel_valid = 0;

    // Apply reset
    printf("Applying Reset\n");
    for (int i = 0; i < 10; i++) {
        top->clk = !top->clk;
        top->eval();
        tfp->dump(main_time);
        main_time++;
    }
    top->rst = 0; // Deassert reset

    // Test image data
    uint8_t test_data[ROWS][IMAGE_SIZE] = {
        {0x01, 0x02, 0x03, 0x04, 0x05},
        {0x06, 0x07, 0x08, 0x09, 0x0A},
        {0x0B, 0x0C, 0x0D, 0x0E, 0x0F},
        {0x10, 0x11, 0x12, 0x13, 0x14},
        {0x15, 0x16, 0x17, 0x18, 0x19}
    };

    printf("Feeding pixel data into the line buffer\n");

    // Feed data row by row
    for (int row = 0; row < ROWS; row++) {
        for (int col = 0; col < IMAGE_SIZE; col++) {
            // Load pixel into pixel_in and set valid
            top->pixel_in = test_data[row][col];
            top->pixel_valid = 1;

            // Toggle clock for one cycle
            for (int i = 0; i < 2; i++) {
                top->clk = !top->clk;
                top->eval();
                tfp->dump(main_time);
                main_time++;
            }

            // Print window contents when valid
            if (top->window_valid) {
                printf("Window at row %d, col %d:\n", row, col);
                for (int i = 0; i < NUM_LINES; i++) {
                    for (int j = 0; j < NUM_LINES; j++) {
                        int idx = (i * NUM_LINES + j);
                        int arrayIndex = (idx * DATA_WIDTH) / 32;
                        int bitOffset = (idx * DATA_WIDTH) % 32;
                        uint32_t value = (top->window[arrayIndex] >> bitOffset) & ((1 << DATA_WIDTH) - 1);
                        printf("%02x ", value);
                    }
                    printf("\n");
                }
                printf("\n");
            }
        }
    }

    // Add additional simulation cycles with valid=0
    printf("Running additional simulation cycles...\n");
    top->pixel_valid = 0;
    for (int i = 0; i < 60; i++) {
        // Toggle clock
        for (int j = 0; j < 2; j++) {
            top->clk = !top->clk;
            top->eval();
            tfp->dump(main_time);
            main_time++;
        }
        
        // Print window contents if valid
        if (top->window_valid) {
            printf("Extra cycle %d - Window Output:\n", i);
            for (int i = 0; i < NUM_LINES; i++) {
                for (int j = 0; j < NUM_LINES; j++) {
                    int idx = (i * NUM_LINES + j);
                    int arrayIndex = (idx * DATA_WIDTH) / 32;
                    int bitOffset = (idx * DATA_WIDTH) % 32;
                    uint32_t value = (top->window[arrayIndex] >> bitOffset) & ((1 << DATA_WIDTH) - 1);
                    printf("%02x ", value);
                }
                printf("\n");
            }
            printf("\n");
        }
    }

    printf("Simulation complete. Final time: %lu\n", main_time);

    // Close trace file and clean up
    tfp->close();
    top->final();
    delete top;

    return 0;
}