#include "include/signal_path.h"
#define BLOCK_LEN 1

/* Global variables decleration*/
static int counter=0;
static int mu;

static int leak=2147462173; //0.999 // (1 ? µ?)


int chess_storage(DMB) fir_lms_delay_line[MAX_FIR_COEFFS];
BufferPtrDMB chess_storage(DMB) ptr_fir_lms_delay_line;
BufferPtr ptr_fir_lms_coeffs;

int chess_storage(DMA % (sizeof(long long))) fir_lms_coeffs[MAX_FIR_COEFFS]; // The coefficients for the adaptive filter


#ifdef PLATFORM_GENERIC
    // lpdsp32 functionallity moddeling functions
    accum_t fract_mult(int a, int b){
        long int a_long = a;
        long int b_long = b;
        return (b_long * a_long);
    }
    accum_t to_accum(int a){
        long int a_long = (long int) a;
        return a_long << 31;
    }
    int rnd_saturate(accum_t a){
        return a >> 31;
    }
    int extract_high(accum_t a){
        return a >> 31;
    }
    void lldecompose(unsigned long long l, int* int1, int* int2){
        *int2 = (int)(l >> 32);
        *int1 = (int)(l);
    }
    uint64_t llcompose(int a, int b) {
        uint64_t result = (uint64_t)b; // Assign b to the higher 32 bits of the result
        result <<= 32; // Shift the higher 32 bits to the left
        result |= (uint32_t)a; // Bitwise OR operation with the lower 32 bits of a
        return result;
    }
    // unsigned long long llcompose(int a, int b){
    //     unsigned long long l;
    //     l = a << 32;
    //     l |= b;
    //     return l;
    //}
    int* cyclic_add(int *ptr, int i_pp, int *ptr_start, int buffer_len){
        int *p_ptr=ptr;
        for (int i=0; i < abs(i_pp); i+=1){ // end of buffer wraparound
            if (i_pp > 0){
                p_ptr ++;
                if (p_ptr >= ptr_start + buffer_len){
                    p_ptr=ptr_start;
                }
            }
            else{ // start of buffer wraparound
                p_ptr--;
                if (p_ptr < ptr_start){
                    p_ptr=ptr_start + (buffer_len -1);
                }
            }
        }
        return p_ptr;
    }
#endif

int sig_init_buffer(BufferPtr *buffer, int *buffer_start_add, int length, int max_buffer_len) {
    buffer->buffer_len = length;
    buffer->ptr_start = buffer_start_add;
    buffer->ptr_current = buffer_start_add;
    // initialize delay line with 0
    for (int i = 0; i < length; i++) {
        buffer_start_add[i] = 0;
    }
    if (length<max_buffer_len){
        return 0;
    }
    else{
        return 1;
    }
}

int sig_init_buffer_DMB(BufferPtrDMB chess_storage(DMB) *buffer, int chess_storage(DMB) *buffer_start_add, int length, int max_buffer_len){
    buffer->buffer_len = length;
    buffer->ptr_start = buffer_start_add;
    buffer->ptr_current = buffer_start_add;
    // initialize delay line with 0
    for (int i = 0; i < length; i++) {
        buffer_start_add[i] = 0;
    }
    if (length<max_buffer_len){
        return 0;
    }
    else{
        return 1;
    }
}

void sig_cirular_buffer_ptr_increment(BufferPtr *buffer, int i_incr){
    buffer->ptr_current = cyclic_add(buffer->ptr_current, i_incr, buffer->ptr_start, buffer->buffer_len);
}

void sig_cirular_buffer_ptr_increment_DMB(BufferPtrDMB *buffer, int i_incr){
    buffer->ptr_current = cyclic_add(buffer->ptr_current, i_incr, buffer->ptr_start, buffer->buffer_len);
}

void sig_cirular_buffer_ptr_put_sample(BufferPtr *buffer, int sample){
    *buffer->ptr_current = sample;
    buffer->ptr_current = cyclic_add(buffer->ptr_current, 1, buffer->ptr_start, buffer->buffer_len);
}

void sig_cirular_buffer_ptr_put_sample_DMB(BufferPtrDMB chess_storage(DMB) *buffer, int sample){
    *buffer->ptr_current = sample;
    buffer->ptr_current = cyclic_add(buffer->ptr_current, 1, buffer->ptr_start, buffer->buffer_len);
}

void static inline sig_circular_buffer_ptr_put_block(BufferPtr *buffer, int* block){
    // increment pointer to oldest block
    //buffer->ptr_current = cyclic_add(buffer->ptr_current, BLOCK_LEN, buffer->ptr_start, buffer->buffer_len);
    // load the next block
    for (int i=0; i<BLOCK_LEN;  i+=2){
        buffer->ptr_current[0] = block[i]; // TODO: use llcompose
        buffer->ptr_current[1] = block[i+1];
        buffer->ptr_current = cyclic_add(buffer->ptr_current, 2, buffer->ptr_start, buffer->buffer_len);
    }
}

//Initialisierungsfunktion für Biquad Filter Koeffizienten
void sig_init_preemph_coef(SingleSignalPath *signal, double b0, double b1, double b2, double a1, double a2, int scale_bits) {
    // Wenn b0=1 und Rest 0 -> kein Filter weil effektiv 1*Xn
    if (b0 == 1. && b1 == 0. && b2 == 0. && a1 == 0. && a2 == 0.) {
        signal->preemph_activated = 0;
    }
    else{
        signal->preemph_activated = 1;              // Schreibe Eintrag in Struct
        signal->_preemph_scale_nbits = scale_bits;  // Schreibe Eintrag in Struct - wieviel Bits wird skaliert
        int scale = pow(2, scale_bits) - 1;         //2^n -1 Skalierung
        // Skaliere Koeffizienten zu Interger und schreibe Eintrag in Struct
        signal->b_preemph[0] = b0 * scale;
        signal->b_preemph[1] = b1 * scale;
        signal->b_preemph[2] = b2 * scale;
        signal->b_preemph[3] = a1 * scale;
        signal->b_preemph[4] = a2 * scale;
    }
}

/*Initialization functions - make sure all of them were called to ensure functionality*/
int sig_init_delay(SingleSignalPath *signal, int n_delay) {
    return sig_init_buffer(&signal->delay_buffer, signal->_delay_buffer, n_delay, MAX_DELAY_SAMPS);
}

//Initialisierungsfunktion für Gewichtung
void sig_init_weight(SingleSignalPath *signal, double weight, int scale_nbits) {
    // Wenn Gewichtung 1 -> kein Effekt
    if (weight == 1.) {
        signal->weight_actived = 0;
    }
    // Wenn Gewichtung != 1 -> Zu Integer skalieren und Eintrag in Struct schreiben
    else{
        signal->weight_actived = 1;
        int scale = pow(2, scale_nbits) - 1;
        signal->weight = weight * scale;
        signal->_weight_scale_nbits = scale_nbits;
    }
}

/*Calculator functions for the given signal path*/
/*Calculate one biquad filter element*/
int sig_calc_biquad(SingleSignalPath *signal, int x) {
    if (signal->preemph_activated == 0) {
        return x;
    }
    accum_t sum =
        fract_mult(x, signal->b_preemph[0]) + fract_mult(signal->_xd[0], signal->b_preemph[1]) +
        fract_mult(signal->_xd[1], signal->b_preemph[2]) + fract_mult(signal->_yd[0], signal->b_preemph[3]) +
        fract_mult(signal->_yd[1],signal->b_preemph[4]);  
    int y = rnd_saturate(sum << 1);

    
    signal->_xd[1] = signal->_xd[0];
    signal->_xd[0] = x;
    signal->_yd[1] = signal->_yd[0];
    signal->_yd[0] = y;
    return y;
}
int inline sig_get_delayed_sample(SingleSignalPath *signal) {
    return *signal->delay_buffer.ptr_current;
}

int sig_delay_buffer_load_and_get(SingleSignalPath *signal, int x) {
    if (signal->delay_buffer.buffer_len == 0) {
        return x;
    }
    int out = *signal->delay_buffer.ptr_current;
    *signal->delay_buffer.ptr_current = x;
    sig_cirular_buffer_ptr_increment(&signal->delay_buffer, 1);
    return out;
}

int sig_calc_weight(SingleSignalPath *signal, int x) {
    if (signal->weight_actived == 0) {
        return x;
    }
    accum_t acc = fract_mult(x, signal->weight);

    return rnd_saturate(acc);
}

int inline sig_calc_fir_lpdsp32_single(BufferPtrDMB chess_storage(DMB) *ptr_fir_lms_delay_line, BufferPtr *ptr_fir_lms_coeffs){

    // Calculate the fir filter output on x to get the canceller
    int chess_storage(DMB) *p_x0 = ptr_fir_lms_delay_line->ptr_current; // chess_storage(DMB)
    int chess_storage(DMB) *px_start = ptr_fir_lms_delay_line->ptr_start;
    int *p_h = ptr_fir_lms_coeffs->ptr_current;
    int delay_line_len = ptr_fir_lms_delay_line->buffer_len;
    int n_coeff = ptr_fir_lms_coeffs->buffer_len;

    int d0,d1,h0,h1;
    accum_t acc1_A = to_accum(0);
    accum_t acc1_B = to_accum(0);
    accum_t acc1_C;

    // iterate over the coefficients to calculate the filter on x - the canceller
    /* Abschaetzung cycles per 2coefficient:
    dual - load : 1
    dual mac and dual load: 1
    -> 48/2 * 2 = 48 cycles for 48 coefficents
    */
    for (int i=0; i < n_coeff; i+=2) chess_loop_range(1,){
        // Use dual load and dual pointer update
        d0 = *p_x0;
        h0 = *p_h;
        p_h++;
        p_x0 = cyclic_add(p_x0, -1, px_start, delay_line_len);

        d1 = *p_x0;
        h1 = *p_h;
        p_h++;
        p_x0 = cyclic_add(p_x0, -1, px_start, delay_line_len);

        acc1_A+=fract_mult(d0, h0);
        acc1_B+=fract_mult(d1, h1);
        #ifndef LPDSP16
        acc1_A = to_accum(rnd_saturate(acc1_A));
        acc1_B = to_accum(rnd_saturate(acc1_B));
        #endif
        
    }
    // Calculate the output sample
    acc1_C = acc1_A + acc1_B;
    //out32 = rnd_saturate(acc1_A);
    return rnd_saturate(acc1_C);
}

void static inline adapt_coeffs_lpdsp32_single_v1(BufferPtrDMB chess_storage(DMB) *ptr_fir_lms_delay_line, BufferPtr *ptr_fir_lms_coeffs, int out){

    int chess_storage(DMA) *p_h0 = ptr_fir_lms_coeffs->ptr_start; //coeff load pointer
    //int chess_storage(DMA) *p_h1 = ptr_fir_lms_coeffs->ptr_start; //coeff store pointer
    int chess_storage(DMB) *p_x0 = ptr_fir_lms_delay_line->ptr_current; // chess_storage(DMB)
    int chess_storage(DMB) *p_x1 = ptr_fir_lms_delay_line->ptr_current; // chess_storage(DMB)

    p_x1 = cyclic_add(p_x1, -1, ptr_fir_lms_delay_line->ptr_start, ptr_fir_lms_delay_line->buffer_len);

    int prod, x0, x1, h0, h1;
    int chess_storage(DMB) *px_start = ptr_fir_lms_delay_line->ptr_start;
    int delay_line_len = ptr_fir_lms_delay_line->buffer_len;
    int n_coeff = ptr_fir_lms_coeffs->buffer_len;

    accum_t acc_A, acc_B;

    // Calculate the first term of the coefficient adaption
    accum_t acc_C = fract_mult(mu, out);
    prod = rnd_saturate(acc_C);

    /* Abschätzung cycles per 2 coefficient:
    dual load coeffs: 1
    single load tab value: 2
    dual mac: 1
    dual rnd_sat - store: 1
    load/store hazard nop: 1
    */
    for (int i=0; i< n_coeff; i+=2) chess_loop_range(1,){
        // Calculate the coefficient wise adaption
        #ifdef PLATFORM_GENERIC
            lldecompose(*((long long *)p_h0), &h0, &h1);
        #else
            lldecompose(*((long long *)p_h0), h0, h1);
        #endif

        acc_A = to_accum(h0);
        acc_B = to_accum(h1);
        
        acc_A += fract_mult(prod, *p_x0); // TODO: This could be further optimized by using all 4 available accums?
        acc_B += fract_mult(prod, *p_x1);
         
        p_x0 = cyclic_add(p_x0, -2, px_start, delay_line_len);
        p_x1 = cyclic_add(p_x1, -2, px_start, delay_line_len);

        // update the current filter coefficients - dual sat; dual store
        *((long long *)p_h0) = llcompose(rnd_saturate(acc_A), rnd_saturate(acc_B));//load/store hazard ! - 1 nop is needed
        p_h0+=2;
    }
}

void init(
    SingleSignalPath *cSensorSignal,
    SingleSignalPath *accSensorSignal,
    double *b_c,
    double *b_acc,
    int delay_c,
    int delay_acc,
    double weight_c,
    double weight_acc,
    double lms_mu,
    int lms_fir_num_coeffs
    ){
    int scale_bits=31;

    // C-Sensor Initialisierung: Biquad, Delay, Weight skalieren und in Struct schreiben
    sig_init_preemph_coef(cSensorSignal, b_c[0], b_c[1], b_c[2], b_c[3], b_c[4], scale_bits);
    sig_init_delay(cSensorSignal, delay_c);
    sig_init_weight(cSensorSignal, weight_c, scale_bits);

    // Acc-Sensor Initialisierung: Biquad, Delay, Weight skalieren und in Struct schreiben
    sig_init_preemph_coef(accSensorSignal, b_acc[0], b_acc[1], b_acc[2], b_acc[3], b_acc[4], scale_bits);
    sig_init_delay(accSensorSignal, delay_acc);
    sig_init_weight(accSensorSignal, weight_acc, 31);

    //Mu Skalierung und in globale Variable schreiben
    int scale = pow(2, scale_bits) - 1;
    mu = lms_mu * scale;
    // Buffer Initialisierung (Delay Line und Koeffizienten) und anschließend alle Werte auf 0 setzen
    sig_init_buffer_DMB(&ptr_fir_lms_delay_line, fir_lms_delay_line, lms_fir_num_coeffs, MAX_FIR_COEFFS);
    sig_init_buffer(&ptr_fir_lms_coeffs, fir_lms_coeffs, lms_fir_num_coeffs, MAX_FIR_COEFFS);

    for (int i = 0; i < lms_fir_num_coeffs; i++) {
        ptr_fir_lms_delay_line.ptr_start[i] = 0;
        ptr_fir_lms_coeffs.ptr_start[i] = 0;
    }
}

// Data d(cSensor) is signal + noise
// x (accSensor) is reference noise signal
void calc(
    SingleSignalPath *cSensorSignal,
    SingleSignalPath *accSensorSignal,
    OutputMode output_mode,
    int16_t volatile chess_storage(DMB) *cSensor,
    int16_t volatile chess_storage(DMB) *accSensor,
    int16_t volatile chess_storage(DMB) *out_16 
    
    ){
    static int chess_storage(DMA) c_block_pre[BLOCK_LEN];
    static int chess_storage(DMA) acc_block_pre[BLOCK_LEN];
    static int chess_storage(DMA) cSensor_32[BLOCK_LEN];
    static int chess_storage(DMA) accSensor_32[BLOCK_LEN];
    
    static int chess_storage(DMB) acc_block_filt[BLOCK_LEN];
    static int chess_storage(DMB) out_32[BLOCK_LEN];

    static int chess_storage(DMA) *p_c_block_pre =c_block_pre;
    static int chess_storage(DMA) *p_acc_block_filt =acc_block_pre;
    static int chess_storage(DMB) *p_out_32=out_32;


    for (uint32_t i=0; i<BLOCK_LEN; i++) chess_loop_range(1,){
        cSensor_32[i] =  ((int) cSensor[i]) << BITSHIFT_16_TO_32;
        accSensor_32[i] = ((int) accSensor[i]) << BITSHIFT_16_TO_32;
    }    

    // Apply bitshift, calculate the pre emphasis filter, delay and weight to each channel     
    for (uint32_t i=0; i<BLOCK_LEN; i++) chess_loop_range(1,){
         c_block_pre[i] = cSensor_32[i];
         acc_block_pre[i] = accSensor_32[i];
    }

    // apply lms filter on cSensor signal
    // Increment the buffer pointer and set the current sample to the delay line
    sig_cirular_buffer_ptr_put_sample_DMB(&ptr_fir_lms_delay_line, acc_block_pre[0]);
    // Calculate the fir filter output on acc to get the canceller
    acc_block_filt[0]= sig_calc_fir_lpdsp32_single(&ptr_fir_lms_delay_line, &ptr_fir_lms_coeffs);
    // Calculate the ouptut signal by c_block_pre - acc_block_filt
    out_32[0] = c_block_pre[0] - acc_block_filt[0];
    // Calculate the coefficient adaptation
    adapt_coeffs_lpdsp32_single_v1(&ptr_fir_lms_delay_line, &ptr_fir_lms_coeffs, out_32[0]);

    // TODO: Add a couple of biqads after ANC
    for (uint32_t i=0; i<BLOCK_LEN; i++) chess_flatten_loop
    {
        out_16[i] = rnd_saturate(to_accum(out_32[i]) >> BITSHIFT_16_TO_32); // 12 cycles for blocksize 4 //TODO: use rnd_saturate(out_32[i] >> input_nbit_bitshift)
    }

}