#include <stdio.h> 
#include <string.h>
#include <math.h>

#include <common_func.h>
#include <confreg_time.h>
#include <fft.h>

// BSP板级支持包所需全局变量
unsigned long UART_BASE = 0xbf000000;
unsigned long CONFREG_TIMER_BASE = 0xbf20f100;
unsigned long CONFREG_CLOCKS_PER_SEC = 50000000L;
unsigned long CORE_CLOCKS_PER_SEC = 33000000L;

const float PI = 3.14159265358979323846;

// 读取定时器的当前Tick
unsigned int get_timer_ticks() {
    return RegRead(CONFREG_TIMER_BASE);
}

// 软件FFT实现 (基2 DIT-FFT 算法)
void sw_fft(float re[], float im[], int n) {
    int i, j, k, l;
    float tr, ti, ur, ui, wr, wi;

    // 比特翻转 (Bit Reversal)
    j = 0;
    for (i = 0; i < n - 1; i++) {
        if (i < j) {
            tr = re[i]; ti = im[i];
            re[i] = re[j]; im[i] = im[j];
            re[j] = tr; im[j] = ti;
        }
        k = n / 2;
        while (k <= j) {
            j -= k;
            k /= 2;
        }
        j += k;
    }

    // 蝶形运算 (Butterfly Computation)
    for (l = 1; l < n; l *= 2) {
        ur = 1.0; 
        ui = 0.0;
        wr = cos(PI / l);
        wi = -sin(PI / l);
        
        for (i = 0; i < n; i += 2 * l) {
            ur = 1.0; 
            ui = 0.0;
            for (j = 0; j < l; j++) {
                int p = i + j;
                int q = p + l;
                
                tr = re[q] * ur - im[q] * ui;
                ti = re[q] * ui + im[q] * ur;
                
                re[q] = re[p] - tr;
                im[q] = im[p] - ti;
                re[p] += tr;
                im[p] += ti;
                
                float next_ur = ur * wr - ui * wi;
                ui = ur * wi + ui * wr;
                ur = next_ur;
            }
        }
    }
}

int main(int argc, char** argv)
{
    unsigned int fft_csr = RegRead(FFT_CSR_REG);
    printf("fft_csr = %x\n", fft_csr);

    // 准备测试数据
    float sw_in_re[FFT_POINT_NUM];
    float sw_in_im[FFT_POINT_NUM];

    for (int i = 0; i < FFT_POINT_NUM; i++) {
        float dc_part   = 4000.0f;
        float f10_part  = 8000.0f * cos(2 * PI * 10.0 * i / FFT_POINT_NUM);
        float f200_part = 6000.0f * sin(2 * PI * 200.0 * i / FFT_POINT_NUM);
        float f400_part = 3000.0f * sin(2 * PI * 400.0 * i / FFT_POINT_NUM);
        
        sw_in_re[i] = dc_part + f10_part + f200_part + f400_part;
        sw_in_im[i] = 0.0f;
    }

    unsigned int tick_start, tick_end;
    unsigned int hw_time, sw_time;

    // 硬件加速 FFT 测试
    printf("\n--- Starting Hardware FFT ---\n");
    tick_start = get_ns();

    for (int i = 0; i < FFT_POINT_NUM; i++) {
        RegWrite(FFT_IN_RE_BASE + (i * 4), (unsigned int)((int)sw_in_re[i]));
        RegWrite(FFT_IN_IM_BASE + (i * 4), 0);
    }
    
    
    fft_start();
    fft_wait();
    
    tick_end = get_ns();
    
    // 如果定时器是递减的，这里可能是 tick_start - tick_end
    // 假设定时器是向上递增的：
    hw_time = tick_end - tick_start;  

    int hw_out_re[FFT_POINT_NUM];
    int hw_out_im[FFT_POINT_NUM];
    for (int i = 0; i < FFT_POINT_NUM; i++) {
        hw_out_re[i] = (int)RegRead(FFT_OUT_RE_BASE + (i * 4));
        hw_out_im[i] = (int)RegRead(FFT_OUT_IM_BASE + (i * 4));
    }

    // ==========================================
    // 2. 纯软件 FFT 测试
    // ==========================================
    printf("--- Starting Software FFT ---\n");
    tick_start = get_ns();
    
    sw_fft(sw_in_re, sw_in_im, FFT_POINT_NUM);
    
    tick_end = get_ns();
    sw_time = tick_end - tick_start;


    // ==========================================
    // 3. 打印对比结果
    // ==========================================
    printf("\n--- Performance Comparison ---\n");
    printf("Timer Clock Freq : %lu Hz\n", CONFREG_CLOCKS_PER_SEC);
    printf("Hardware FFT Time: %u ns (%.3f ms)\n", hw_time, (float)hw_time / 1000000.0);
    printf("Software FFT Time: %u ns (%.3f ms)\n", sw_time, (float)sw_time / 1000000.0);
    
    if (hw_time > 0) {
        printf("Speedup Ratio    : %.2fx\n", (float)sw_time / hw_time);
    }

    printf("\n--- Verification (Only showing Bins with energy > 10) ---\n");
    for (int i = 0; i < FFT_POINT_NUM; i++) {
        if (fabs(hw_out_re[i]) > 10 || fabs(hw_out_im[i]) > 10) {
            printf("Bin [%4d] Hz: HW(Re:%6d, Im:%6d) | SW(Re:%6d, Im:%6d)\n", 
                    i, 
                    hw_out_re[i], hw_out_im[i], 
                    (int)sw_in_re[i] / FFT_POINT_NUM, (int)sw_in_im[i] / FFT_POINT_NUM);
        }
    }

    return 0;
}