#include "synth_engine.h"
#include <math.h>
#include <utility>
#include <string.h>

// A simple sine lookup table for the sine oscillator
const int SINE_TABLE_SIZE = 256;
static int16_t sine_table[SINE_TABLE_SIZE];
static bool sine_table_filled = false;

/**
 * @brief Fills the global sine table. Called once on startup.
 */
void fill_sine_table() {
    if (sine_table_filled) return;
    for (int i = 0; i < SINE_TABLE_SIZE; ++i) {
        // M_PI is not standard C++, but it's common. If it fails, use 3.1415926535...
        sine_table[i] = static_cast<int16_t>(sin(2.0 * M_PI * i / SINE_TABLE_SIZE) * 32767.0);
    }
    sine_table_filled = true;
}

SynthEngine::SynthEngine(uint32_t sampleRate)
    : grid{},
      _sampleRate(sampleRate),
      _phase(0),
      _increment(0),
      _volume(0.5f),
      _waveform(SAWTOOTH),
      _isGateOpen(false),
      _freqToPhaseInc(0.0f),
      _rngState(12345)
{
    fill_sine_table();
    // Initialize with a default frequency
    setFrequency(440.0f);

    // Initialize SINK
    _freqToPhaseInc = 4294967296.0f / (float)_sampleRate;
    grid[GRID_W / 2][GRID_H - 1].type = GridCell::SINK;
    rebuildProcessingOrder();
}

SynthEngine::~SynthEngine() {
}

size_t SynthEngine::exportGrid(uint8_t* buffer) {
    SynthLockGuard<SynthMutex> lock(gridMutex);
    uint8_t count = 0;
    for(int y=0; y<GRID_H; ++y) {
        for(int x=0; x<GRID_W; ++x) {
            if (grid[x][y].type != GridCell::EMPTY) count++;
        }
    }

    size_t idx = 0;
    buffer[idx++] = count;

    for(int y=0; y<GRID_H; ++y) {
        for(int x=0; x<GRID_W; ++x) {
            GridCell& c = grid[x][y];
            if (c.type != GridCell::EMPTY) {
                buffer[idx++] = (uint8_t)x;
                buffer[idx++] = (uint8_t)y;
                buffer[idx++] = (uint8_t)c.type;
                buffer[idx++] = (uint8_t)((c.param * 255) >> FP_SHIFT);
                buffer[idx++] = (uint8_t)c.rotation;
            }
        }
    }
    buffer[idx++] = count;
    return idx;
}

int SynthEngine::importGrid(const uint8_t* buffer, size_t size) {
    if (size < 2) return 1;
    uint8_t countStart = buffer[0];
    uint8_t countEnd = buffer[size - 1];
    
    if (countStart != countEnd) return 2;
    
    size_t expectedSize = 1 + countStart * 5 + 1;
    if (size != expectedSize) return 3;

    SynthLockGuard<SynthMutex> lock(gridMutex);

    // Clear grid first
    for (int x = 0; x < GRID_W; ++x) {
        for (int y = 0; y < GRID_H; ++y) {
            GridCell& c = grid[x][y];
            if (c.type == GridCell::SINK) continue;
            c.type = GridCell::EMPTY;
            c.param = FP_HALF;
            c.rotation = 0;
            c.value = 0;
            c.phase = 0;
            c.phase_accumulator = 0;
            c.next_value = 0;
        }
    }

    size_t idx = 1;
    for(int i=0; i<countStart; ++i) {
        uint8_t x = buffer[idx++];
        uint8_t y = buffer[idx++];
        uint8_t t = buffer[idx++];
        uint8_t p = buffer[idx++];
        uint8_t r = buffer[idx++];
        
        if (x < GRID_W && y < GRID_H) {
            GridCell& c = grid[x][y];
            c.type = (GridCell::Type)t;
            c.param = ((int32_t)p << FP_SHIFT) / 255;
            c.rotation = r;
        }
    }
    rebuildProcessingOrder_locked();
    return 0;
}

void SynthEngine::clearGrid() {
    SynthLockGuard<SynthMutex> lock(gridMutex);
    for (int x = 0; x < GRID_W; ++x) {
        for (int y = 0; y < GRID_H; ++y) {
            GridCell& c = grid[x][y];
            if (c.type == GridCell::SINK) continue;

            c.type = GridCell::EMPTY;
            c.param = FP_HALF;
            c.rotation = 0;
            c.value = 0;
            c.phase = 0;
            c.phase_accumulator = 0;
            c.next_value = 0;
        }
    }
    rebuildProcessingOrder_locked();
}

void SynthEngine::loadPreset(int preset) {
    clearGrid();
    SynthLockGuard<SynthMutex> lock(gridMutex);

    auto placeOp = [&](int x, int y, float ratio, float att, float rel) {
        // Layout:
        // (x, y)   : G-IN (South)
        // (x, y+1) : WIRE (East) -> Feeds envelope chain
        // (x+1, y+1): ATT (East) ->
        // (x+2, y+1): REL (East)
        // (x+3, y+1): VCA (South) -> Output is here. Gets audio from OSC, gain from envelope.
        // (x+3, y) : OSC (South) -> Audio source. Gets FM from its back (x+3, y-1).

        grid[x][y].type = GridCell::GATE_INPUT; grid[x][y].rotation = 2; // S
        grid[x][y+1].type = GridCell::WIRE; grid[x][y+1].rotation = 1; // E

        grid[x+1][y+1].type = GridCell::ADSR_ATTACK; grid[x+1][y+1].rotation = 1; // E
        grid[x+1][y+1].param = (int32_t)(att * FP_ONE);

        grid[x+2][y+1].type = GridCell::ADSR_RELEASE; grid[x+2][y+1].rotation = 1; // E
        grid[x+2][y+1].param = (int32_t)(rel * FP_ONE);

        grid[x+3][y+1].type = GridCell::VCA; grid[x+3][y+1].rotation = 2; // S
        grid[x+3][y+1].param = 0; // Controlled by Env

        grid[x+3][y].type = GridCell::INPUT_OSCILLATOR; grid[x+3][y].rotation = 2; // S
        grid[x+3][y].param = (ratio > 1.0f) ? FP_HALF : 0;
    };

    int sinkY = GRID_H - 1;
    int sinkX = GRID_W / 2;

    if (preset == 1) { // Based on DX7 Algorithm 32
        placeOp(0, 0, 1.0f, 0.01f, 0.5f); // Op 1
        placeOp(4, 0, 1.0f, 0.05f, 0.3f); // Op 2
        placeOp(8, 0, 2.0f, 0.01f, 0.2f); // Op 3
        
        grid[3][2].type = GridCell::WIRE; grid[3][2].rotation = 2;
        grid[3][3].type = GridCell::WIRE; grid[3][3].rotation = 1; // E
        grid[4][3].type = GridCell::WIRE; grid[4][3].rotation = 1; // E
        grid[5][3].type = GridCell::WIRE; grid[5][3].rotation = 1; // E
        grid[6][3].type = GridCell::WIRE; grid[6][3].rotation = 2; // S
        
        grid[7][2].type = GridCell::WIRE; grid[7][2].rotation = 2;
        grid[7][3].type = GridCell::WIRE; grid[7][3].rotation = 3; // W
        
        grid[11][2].type = GridCell::WIRE; grid[11][2].rotation = 2;
        grid[11][3].type = GridCell::WIRE; grid[11][3].rotation = 3; // W
        grid[10][3].type = GridCell::WIRE; grid[10][3].rotation = 3; // W
        grid[9][3].type = GridCell::WIRE; grid[9][3].rotation = 3; // W
        grid[8][3].type = GridCell::WIRE; grid[8][3].rotation = 3; // W

        for(int y=4; y<sinkY; ++y) {
            grid[6][y].type = GridCell::WIRE; grid[6][y].rotation = 2;
        }
    } else if (preset == 2) { // Algo 1: Stack (FM)
        placeOp(4, 0, 2.0f, 0.01f, 0.2f); // Modulator
        placeOp(4, 2, 1.0f, 0.01f, 0.8f); // Carrier
        
        grid[7][4].type = GridCell::WIRE; grid[7][4].rotation = 3; // W
        grid[6][4].type = GridCell::WIRE; grid[6][4].rotation = 2; // S
        for(int y=5; y<sinkY; ++y) {
            grid[6][y].type = GridCell::WIRE; grid[6][y].rotation = 2;
        }
    } else if (preset == 3) { // Algo 2
        placeOp(4, 2, 1.0f, 0.01f, 0.8f);
        placeOp(4, 0, 2.0f, 0.01f, 0.2f);
        placeOp(0, 0, 1.0f, 0.01f, 0.5f);

        grid[7][4].type = GridCell::WIRE; grid[7][4].rotation = 3; // W
        
        grid[3][2].type = GridCell::WIRE; grid[3][2].rotation = 1; // E
        grid[4][2].type = GridCell::WIRE; grid[4][2].rotation = 1; // E
        grid[5][2].type = GridCell::WIRE; grid[5][2].rotation = 1; // E
        grid[sinkX][2].type = GridCell::WIRE; grid[sinkX][2].rotation = 2; // S
        grid[sinkX][3].type = GridCell::WIRE; grid[sinkX][3].rotation = 2; // S
        grid[sinkX][4].type = GridCell::WIRE; grid[sinkX][4].rotation = 2; // S

        for(int y=5; y<sinkY; ++y) {
            grid[sinkX][y].type = GridCell::WIRE; grid[sinkX][y].rotation = 2;
        }
    } else if (preset == 4) { // Algo 4
        placeOp(4, 4, 1.0f, 0.01f, 0.8f); 
        placeOp(4, 2, 2.0f, 0.01f, 0.2f); 
        placeOp(4, 0, 4.0f, 0.01f, 0.1f); 

        grid[7][6].type = GridCell::WIRE; grid[7][6].rotation = 3; // W
        grid[sinkX][6].type = GridCell::WIRE; grid[sinkX][6].rotation = 2; // S

        for(int y=7; y<sinkY; ++y) {
            grid[sinkX][y].type = GridCell::WIRE; grid[sinkX][y].rotation = 2;
        }
    } else if (preset == 5) { // Algo 5
        placeOp(4, 4, 1.0f, 0.01f, 0.8f);
        placeOp(4, 2, 2.0f, 0.01f, 0.2f);
        placeOp(4, 0, 4.0f, 0.01f, 0.1f);
        placeOp(0, 0, 0.5f, 0.01f, 0.5f);

        grid[7][6].type = GridCell::WIRE; grid[7][6].rotation = 3; // W
        grid[3][2].type = GridCell::WIRE; grid[3][2].rotation = 2; // S
        grid[3][3].type = GridCell::WIRE; grid[3][3].rotation = 1; // E
        grid[4][3].type = GridCell::WIRE; grid[4][3].rotation = 1; // E
        grid[5][3].type = GridCell::WIRE; grid[5][3].rotation = 1; // E
        grid[6][3].type = GridCell::WIRE; grid[6][3].rotation = 2; // S
        grid[sinkX][4].type = GridCell::WIRE; grid[sinkX][4].rotation = 2; // S
        grid[sinkX][5].type = GridCell::WIRE; grid[sinkX][5].rotation = 2; // S
        
        grid[sinkX][6].type = GridCell::WIRE; grid[sinkX][6].rotation = 2; // S

        for(int y=7; y<sinkY; ++y) {
            grid[sinkX][y].type = GridCell::WIRE; grid[sinkX][y].rotation = 2;
        }
    }

    rebuildProcessingOrder_locked();
}

void SynthEngine::setFrequency(float freq) {
    // Calculate the phase increment for a given frequency.
    // The phase accumulator is a 32-bit unsigned integer (0 to 2^32-1).
    // One full cycle of the accumulator represents one cycle of the waveform.
    // increment = (frequency * 2^32) / sampleRate
    // The original calculation was incorrect for float frequencies.
    _increment = static_cast<uint32_t>((double)freq * (4294967296.0 / (double)_sampleRate));
}

void SynthEngine::setVolume(float vol) {
    if (vol < 0.0f) vol = 0.0f;
    if (vol > 1.0f) vol = 1.0f;
    _volume = vol;
}

void SynthEngine::setWaveform(Waveform form) {
    _waveform = form;
}

void SynthEngine::setGate(bool isOpen) {
    _isGateOpen = isOpen;
}

float SynthEngine::getFrequency() const {
    return (float)((double)_increment * (double)_sampleRate / 4294967296.0);
}

int32_t SynthEngine::_random() {
    // Simple Linear Congruential Generator
    _rngState = _rngState * 1664525 + 1013904223;
    return (int32_t)((_rngState >> 16) & 0xFFFF) - 32768;
}

void SynthEngine::rebuildProcessingOrder_locked() {
    _processing_order.clear();
    bool visited[GRID_W][GRID_H] = {false};
    std::vector<std::pair<int, int>> q;

    // Start BFS from the SINK backwards
    q.push_back({GRID_W / 2, GRID_H - 1});
    visited[GRID_W / 2][GRID_H - 1] = true;

    int head = 0;
    while(head < (int)q.size()) {
        std::pair<int, int> curr = q[head++];
        int cx = curr.first;
        int cy = curr.second;
        
        // Check neighbors to see if they output to (cx, cy)
        int nx_offsets[4] = {0, 1, 0, -1};
        int ny_offsets[4] = {-1, 0, 1, 0};
        
        for(int i=0; i<4; ++i) {
            int tx = cx + nx_offsets[i];
            int ty = cy + ny_offsets[i];

            if (tx >= 0 && tx < GRID_W && ty >= 0 && ty < GRID_H && !visited[tx][ty]) {
                GridCell& neighbor = grid[tx][ty];
                bool pointsToCurr = false;
                
                if (neighbor.type != GridCell::EMPTY && neighbor.type != GridCell::SINK) {
                    int dx = cx - tx;
                    int dy = cy - ty;
                    int dir = -1;
                    if (dx == 0 && dy == -1) dir = 0; // N
                    else if (dx == 1 && dy == 0) dir = 1; // E
                    else if (dx == 0 && dy == 1) dir = 2; // S
                    else if (dx == -1 && dy == 0) dir = 3; // W

                    if (neighbor.type == GridCell::FORK) {
                        int leftOut = (neighbor.rotation + 3) % 4;
                        int rightOut = (neighbor.rotation + 1) % 4;
                        if (dir == leftOut || dir == rightOut) pointsToCurr = true;
                    } else {
                        if (neighbor.rotation == dir) pointsToCurr = true;
                    }
                }
                
                if (pointsToCurr) {
                    visited[tx][ty] = true;
                    q.push_back({tx, ty});
                }
            }
        }
    }

    _processing_order = q;
}

void SynthEngine::rebuildProcessingOrder() {
    SynthLockGuard<SynthMutex> lock(gridMutex);
    rebuildProcessingOrder_locked();
}

void SynthEngine::updateGraph() {
    rebuildProcessingOrder_locked();
}

int32_t SynthEngine::processGridStep() {

    auto isConnected = [&](int tx, int ty, int from_x, int from_y) -> bool {
        if (from_x < 0 || from_x >= GRID_W || from_y < 0 || from_y >= GRID_H) return false;
        GridCell& n = grid[from_x][from_y];
        
        bool connects = false;
        if (n.type == GridCell::WIRE || n.type == GridCell::FIXED_OSCILLATOR || n.type == GridCell::INPUT_OSCILLATOR || n.type == GridCell::WAVETABLE || n.type == GridCell::NOISE || n.type == GridCell::LFO || n.type == GridCell::GATE_INPUT || n.type == GridCell::ADSR_ATTACK || n.type == GridCell::ADSR_DECAY || n.type == GridCell::ADSR_SUSTAIN || n.type == GridCell::ADSR_RELEASE || n.type == GridCell::LPF || n.type == GridCell::HPF || n.type == GridCell::VCA || n.type == GridCell::BITCRUSHER || n.type == GridCell::DISTORTION || n.type == GridCell::RECTIFIER || n.type == GridCell::PITCH_SHIFTER || n.type == GridCell::GLITCH || n.type == GridCell::OPERATOR || n.type == GridCell::DELAY || n.type == GridCell::REVERB) {
            // Check rotation
            // 0:N (y-1), 1:E (x+1), 2:S (y+1), 3:W (x-1)
            if (n.rotation == 0 && from_y - 1 == ty && from_x == tx) connects = true;
            if (n.rotation == 1 && from_x + 1 == tx && from_y == ty) connects = true;
            if (n.rotation == 2 && from_y + 1 == ty && from_x == tx) connects = true;
            if (n.rotation == 3 && from_x - 1 == tx && from_y == ty) connects = true;
        } else if (n.type == GridCell::FORK) {
            // Fork outputs to Left (rot+3) and Right (rot+1) relative to its rotation
            // n.rotation is "Forward"
            int dx = tx - from_x;
            int dy = ty - from_y;
            int dir = -1;
            if (dx == 0 && dy == -1) dir = 0; // N
            if (dx == 1 && dy == 0) dir = 1; // E
            if (dx == 0 && dy == 1) dir = 2; // S
            if (dx == -1 && dy == 0) dir = 3; // W
            
            int leftOut = (n.rotation + 3) % 4;
            int rightOut = (n.rotation + 1) % 4;
            
            if (dir == leftOut || dir == rightOut) connects = true;
        }
        return connects;
    };

    // Helper to get input from a neighbor
    auto getInput = [&](int tx, int ty, int from_x, int from_y) -> int32_t {
        if (!isConnected(tx, ty, from_x, from_y)) return 0;
        GridCell& n = grid[from_x][from_y];

        if (n.type == GridCell::FORK) {
            int dx = tx - from_x;
            int dy = ty - from_y;
            int dir = -1;
            if (dx == 0 && dy == -1) dir = 0; // N
            if (dx == 1 && dy == 0) dir = 1; // E
            if (dx == 0 && dy == 1) dir = 2; // S
            if (dx == -1 && dy == 0) dir = 3; // W
            
            int leftOut = (n.rotation + 3) % 4;
            int rightOut = (n.rotation + 1) % 4;
            
            if (dir == leftOut) return (n.value * (FP_ONE - n.param)) >> (FP_SHIFT - 1);
            if (dir == rightOut) return (n.value * n.param) >> (FP_SHIFT - 1);
        }
        
        return n.value;
    };

    // Helper to sum inputs excluding the output direction
    auto getSummedInput = [&](int x, int y, GridCell& c) -> int32_t {
        int32_t sum = 0;
        int outDir = c.rotation; // 0:N, 1:E, 2:S, 3:W
        if (outDir != 0) sum += getInput(x, y, x, y-1);
        if (outDir != 1) sum += getInput(x, y, x+1, y);
        if (outDir != 2) sum += getInput(x, y, x, y+1);
        if (outDir != 3) sum += getInput(x, y, x-1, y);
        return sum;
    };

    auto getInputFromTheBack = [&](int x, int y, GridCell& c) -> int32_t {
        int inDir = (c.rotation + 2) % 4;
        int dx=0, dy=0;
        if(inDir==0) dy=-1; else if(inDir==1) dx=1; else if(inDir==2) dy=1; else dx=-1;
        return getInput(x, y, x+dx, y+dy);
    };

    auto getSideInputGain = [&](int x, int y, GridCell& c) -> int32_t {
        int32_t gain = 0;
        bool hasSide = false;
        // Left (rot+3)
        int lDir = (c.rotation + 3) % 4;
        int ldx=0, ldy=0; if(lDir==0) ldy=-1; else if(lDir==1) ldx=1; else if(lDir==2) ldy=1; else ldx=-1;
        if (isConnected(x, y, x+ldx, y+ldy)) { hasSide = true; gain += getInput(x, y, x+ldx, y+ldy); }
        // Right (rot+1)
        int rDir = (c.rotation + 1) % 4;
        int rdx=0, rdy=0; if(rDir==0) rdy=-1; else if(rDir==1) rdx=1; else if(rDir==2) rdy=1; else rdx=-1;
        if (isConnected(x, y, x+rdx, y+rdy)) { hasSide = true; gain += getInput(x, y, x+rdx, y+rdy); }
        return hasSide ? gain : FP_ONE;
    };

    // 1. Calculate next values for active cells
    for (const auto& cell_coord : _processing_order) {
        int x = cell_coord.first;
        int y = cell_coord.second;
        GridCell& c = grid[x][y];
        int32_t val = 0;

        if (c.type == GridCell::EMPTY) {
            val = 0;
        } else if (c.type == GridCell::FIXED_OSCILLATOR) {
            // Gather inputs for modulation
            int32_t mod = getInputFromTheBack(x, y, c);

            // Freq 10 to 1000 Hz. 
            int32_t freq = 10 + ((c.param * 990) >> FP_SHIFT) + ((mod * 500) >> FP_SHIFT);
            if (freq < 1) freq = 1;
            
            // Fixed point phase accumulation
            uint32_t inc = freq * 97391;
            c.phase_accumulator += inc;
            // Top 8 bits of 32-bit accumulator form the 256-entry table index
            val = sine_table[c.phase_accumulator >> 24];
            val = (val * getSideInputGain(x, y, c)) >> FP_SHIFT;
        } else if (c.type == GridCell::INPUT_OSCILLATOR) {
            int32_t mod = getInputFromTheBack(x, y, c);

            // Freq based on current note + octave param (1-5)
            int octave = 1 + ((c.param * 5) >> FP_SHIFT); // Map 0.0-1.0 to 1-5
            
            // Use the engine's global increment directly to avoid float conversion round-trip
            uint32_t baseInc = _increment;
            uint32_t inc = baseInc << (octave - 1);
            
            // Apply FM (mod is float, convert to fixed point increment)
            inc += (int32_t)(((int64_t)mod * 500 * 97391) >> FP_SHIFT);
            
            c.phase_accumulator += inc;
            val = sine_table[c.phase_accumulator >> 24];
            val = (val * getSideInputGain(x, y, c)) >> FP_SHIFT;
        } else if (c.type == GridCell::WAVETABLE) {
            int32_t mod = getInputFromTheBack(x, y, c);

            // Track current note frequency + FM. Use direct increment for speed.
            uint32_t inc = _increment + (int32_t)(((int64_t)mod * 500 * 97391) >> FP_SHIFT);
            c.phase_accumulator += inc;

            int wave_select = (c.param * 8) >> FP_SHIFT;
            bool phase_upper = (c.phase_accumulator & 0x80000000);

            switch(wave_select) {
                case 0: val = sine_table[c.phase_accumulator >> 24]; break;
                case 1: val = (int32_t)((c.phase_accumulator >> 16) & 0xFFFF) - 32768; break; // Saw
                case 2: val = phase_upper ? -32767 : 32767; break; // Square
                case 3: val = sine_table[c.phase_accumulator >> 24]; break; // Triangle (fallback)
                case 4: val = 32767 - (int32_t)((c.phase_accumulator >> 16) & 0xFFFF); break; // Ramp
                case 5: val = ((c.phase_accumulator >> 30) == 0) ? 32767 : -32767; break; // Pulse 25%
                default:
                    val = sine_table[c.phase_accumulator >> 24];
                    break;
            }
            val = (val * getSideInputGain(x, y, c)) >> FP_SHIFT;
        } else if (c.type == GridCell::NOISE) {
            int32_t mod = getInputFromTheBack(x, y, c);

            int32_t white = _random();
            int shade = (c.param * 5) >> FP_SHIFT;
            switch(shade) {
                case 0: // Brown (Leaky integrator)
                    c.phase = (c.phase + (white >> 3)) - (c.phase >> 4);
                    val = c.phase * 3; // Gain up
                    break;
                case 1: // Pink (Approx: LPF)
                    c.phase = (c.phase >> 1) + (white >> 1);
                    val = c.phase;
                    break;
                case 2: // White
                    val = white;
                    break;
                case 3: // Yellow (HPF)
                    val = white - c.phase;
                    c.phase = white; // Store last sample
                    break;
                case 4: // Green (BPF approx)
                    c.phase = (c.phase + white) >> 1; // LPF
                    val = white - c.phase; // HPF result
                    break;
            }
            
            // Apply Amplitude Modulation (AM) from input
            val = (val * (FP_ONE + mod)) >> FP_SHIFT;
            val = (val * getSideInputGain(x, y, c)) >> FP_SHIFT;
        } else if (c.type == GridCell::LFO) {
            // Low Frequency Oscillator (0.1 Hz to 20 Hz)
            int32_t freq_x10 = 1 + ((c.param * 199) >> FP_SHIFT);
            uint32_t inc = freq_x10 * 9739;
            c.phase_accumulator += inc;
            // Output full range -1.0 to 1.0
            val = sine_table[c.phase_accumulator >> 24];
        } else if (c.type == GridCell::FORK) {
            // Sum inputs from "Back" (Input direction)
            val = getInputFromTheBack(x, y, c);
        } else if (c.type == GridCell::GATE_INPUT) {
            // Outputs 1.0 when gate is open (key pressed), 0.0 otherwise
            val = _isGateOpen ? FP_MAX : 0;
        } else if (c.type == GridCell::ADSR_ATTACK) {
            // Slew Limiter (Up only)
            int32_t in = getInputFromTheBack(x, y, c);
            int32_t rate = (1 << 20) / (1 + (c.param >> 4));
            if (in > (c.phase >> 9)) {
                c.phase += rate;
                if ((c.phase >> 9) > in) c.phase = in << 9;
            } else {
                c.phase = in << 9;
            }
            val = c.phase >> 9;
        } else if (c.type == GridCell::ADSR_DECAY || c.type == GridCell::ADSR_RELEASE) {
            // Slew Limiter (Down only)
            int32_t in = getInputFromTheBack(x, y, c);
            int32_t rate = (1 << 20) / (1 + (c.param >> 4));
            if (in < (c.phase >> 9)) {
                c.phase -= rate;
                if ((c.phase >> 9) < in) c.phase = in << 9;
            } else {
                c.phase = in << 9;
            }
            val = c.phase >> 9;
        } else if (c.type == GridCell::ADSR_SUSTAIN) {
            // Attenuator
            int32_t in = getInputFromTheBack(x, y, c);
            val = (in * c.param) >> FP_SHIFT;
        } else if (c.type == GridCell::WIRE) {
            // Sum inputs from all neighbors that point to me
            val = getSummedInput(x, y, c);
        } else if (c.type == GridCell::LPF) {
            // Input from Back
            int32_t in = getInputFromTheBack(x, y, c);
            
            // Simple one-pole LPF
            int32_t alpha = (c.param * c.param) >> FP_SHIFT;
            
            // c.phase stores previous output
            val = c.phase + ((alpha * (in - c.phase)) >> FP_SHIFT);
            c.phase = val;
        } else if (c.type == GridCell::HPF) {
            // Input from Back
            int32_t in = getInputFromTheBack(x, y, c);
            
            int32_t alpha = (c.param * c.param) >> FP_SHIFT;
            
            // HPF = Input - LPF
            int32_t lpf = c.phase + ((alpha * (in - c.phase)) >> FP_SHIFT);
            c.phase = lpf;
            val = in - lpf;
        } else if (c.type == GridCell::VCA) {
            // Input from Back
            int32_t in = getInputFromTheBack(x, y, c);
            
            // Mod from other directions (sum)
            int32_t mod = getSummedInput(x, y, c);
            mod -= in; // Remove signal input from mod sum (it was included in getInput calls)
            
            // Gain = Param + Mod
            int32_t gain = c.param + mod;
            if (gain < 0) gain = 0;
            val = (in * gain) >> FP_SHIFT;
        } else if (c.type == GridCell::BITCRUSHER) {
            int32_t in = getInputFromTheBack(x, y, c);
            
            // Bit depth reduction
            int32_t mask = 0xFFFF << (16 - (c.param >> 11));
            val = in & mask;
        } else if (c.type == GridCell::DISTORTION) {
            int32_t in = getInputFromTheBack(x, y, c);
            
            // Soft clipping
            int32_t drive = FP_ONE + (c.param << 2);
            val = (in * drive) >> FP_SHIFT;
            if (val > FP_MAX) val = FP_MAX;
            if (val < FP_MIN) val = FP_MIN;
        } else if (c.type == GridCell::RECTIFIER) {
            int32_t in = getInputFromTheBack(x, y, c);
            // Mix between original and rectified based on param
            int32_t rect = (in < 0) ? -in : in;
            val = ((in * (FP_ONE - c.param)) >> FP_SHIFT) + ((rect * c.param) >> FP_SHIFT);
        } else if (c.type == GridCell::GLITCH) {
            int32_t in = getInputFromTheBack(x, y, c);
            // Param controls probability of glitch
            int32_t chance = c.param >> 2;
            if ((_random() & 0x7FFF) < chance) {
                int mode = _random() & 3;
                if (mode == 0) val = in << 4; // Massive gain (clipping)
                else if (mode == 1) val = _random(); // White noise burst
                else val = 0; // Drop out
            } else {
                val = in;
            }
        } else if (c.type == GridCell::OPERATOR || c.type == GridCell::SINK) {
            // Gather inputs
            int32_t inputs[4];
            int count = 0;
            int outDir = (c.type == GridCell::SINK) ? -1 : c.rotation;

            int32_t iN = (outDir != 0) ? getInput(x, y, x, y-1) : 0; if(iN!=0) inputs[count++] = iN;
            int32_t iE = (outDir != 1) ? getInput(x, y, x+1, y) : 0; if(iE!=0) inputs[count++] = iE;
            int32_t iS = (outDir != 2) ? getInput(x, y, x, y+1) : 0; if(iS!=0) inputs[count++] = iS;
            int32_t iW = (outDir != 3) ? getInput(x, y, x-1, y) : 0; if(iW!=0) inputs[count++] = iW;

            if (c.type == GridCell::SINK) {
                // Sink just sums everything
                val = 0;
                for(int k=0; k<count; ++k) val += inputs[k];
            } else {
                // Operator
                int opType = (c.param * 6) >> FP_SHIFT;
                if (count == 0) val = 0;
                else {
                    val = inputs[0];
                    for (int i=1; i<count; ++i) {
                        switch(opType) {
                            case 0: val += inputs[i]; break; // ADD
                            case 1: val = (val * inputs[i]) >> FP_SHIFT; break; // MUL
                            case 2: val -= inputs[i]; break; // SUB
                            case 3: if(inputs[i]!=0) val = (val << FP_SHIFT) / inputs[i]; break; // DIV
                            case 4: if(inputs[i]<val) val = inputs[i]; break; // MIN
                            case 5: if(inputs[i]>val) val = inputs[i]; break; // MAX
                        }
                    }
                }
            }
        } // End of big switch
        c.next_value = val;
    } // End of for loop over _processing_order

    // 2. Update current values from next values for active cells
    for (const auto& cell_coord : _processing_order) {
        int x = cell_coord.first;
        int y = cell_coord.second;
        grid[x][y].value = grid[x][y].next_value;
    }

    return grid[GRID_W / 2][GRID_H - 1].value;
}

void SynthEngine::process(int16_t* buffer, uint32_t numFrames) {
    // Lock grid mutex to prevent UI from changing grid structure mid-process
    SynthLockGuard<SynthMutex> lock(gridMutex);

    for (uint32_t i = 0; i < numFrames; ++i) {
        // The grid is now the primary sound source.
        // The processGridStep() returns Q15
        int32_t sample = processGridStep();

        // Soft clip grid sample to avoid harsh distortion before filtering.
        if (sample > FP_MAX) sample = FP_MAX;
        if (sample < FP_MIN) sample = FP_MIN;

        // The filters were designed for a signal in the int16 range.
        // We scale the grid's output to match this expected range.
        // It is already Q15, so it matches int16 range.

        // Apply Master Volume and write to buffer
        buffer[i] = (int16_t)((sample * (int32_t)(_volume * FP_ONE)) >> FP_SHIFT);
    }
}