//! I2S audio output via PIO + DMA, using the embassy-rp `PioI2sOut` driver.
//!
//! Targets a PCM5102A (or compatible) I2S DAC wired as:
//!   GPIO  9  →  BCK  (bit clock)
//!   GPIO 10  →  LRCK (word select)
//!   GPIO 11  →  DATA
//!
//! # Buffer format
//! `PioI2sOut` expects a flat `&[u32]` of interleaved stereo words:
//! `[L0, R0, L1, R1, …]`.  Each word holds a 24-bit signed sample
//! left-justified in bits 31..8, which is the format expected by PCM5102A in
//! 32-bit I2S mode.
//!
//! # Double-buffering
//! Two static DMA buffers (A and B) alternate:
//!   1. DMA drains buffer A → PIO TX FIFO.
//!   2. CPU renders the next block into buffer B.
//!   3. Swap and repeat.
//!
//! At 48 kHz with 256 frames per block (5.33 ms/block) the Cortex-M33 at
//! 150 MHz has ~10× headroom for the DSP render.

use embassy_executor::task;
use embassy_rp::{
    peripherals::{DMA_CH0, PIN_9, PIN_10, PIN_11, PIO0},
    pio::Pio,
    pio_programs::i2s::{PioI2sOut, PioI2sOutProgram},
    Peri,
};
use synth_core::{
    config::SR_48000,
    envelope::Adsr,
    filter::{FilterMode, Svf},
    math::midi_note_to_hz,
    oscillator::Vco,
    vca::Vca,
    AudioProcessor,
};

use crate::{Irqs, PARAMS};

/// Stereo frames per render block (one frame = left sample + right sample).
const BLOCK_FRAMES: usize = 256;
/// DMA buffer length in u32 words (two words per frame: L and R).
const BLOCK_WORDS: usize = BLOCK_FRAMES * 2;
const SAMPLE_RATE: u32 = 48_000;
const BIT_DEPTH: u32 = 24;

// Ping-pong DMA buffers. Placed in .bss (zero-initialised RAM) at link time.
// SAFETY: only `audio_task` ever accesses these after `main` has run.
static mut DMA_BUF_A: [u32; BLOCK_WORDS] = [0; BLOCK_WORDS];
static mut DMA_BUF_B: [u32; BLOCK_WORDS] = [0; BLOCK_WORDS];

/// Convert a synth-core f32 sample (−1.0 to +1.0) to a 24-bit signed integer
/// left-justified in a u32, as expected by PCM5102A in 32-bit I2S mode.
#[inline]
fn f32_to_i2s(s: f32) -> u32 {
    let clamped = if s > 1.0 { 1.0 } else if s < -1.0 { -1.0 } else { s };
    let i24 = (clamped * 8_388_607.0) as i32;
    // Left-justify: move to bits 31..8
    (i24 << 8) as u32
}

/// Render one `BLOCK_FRAMES`-frame stereo block into `buf` using synth-core DSP.
fn render(
    buf: &mut [u32; BLOCK_WORDS],
    vco: &mut Vco,
    adsr: &mut Adsr,
    filt: &mut Svf,
    vca: &mut Vca,
    snap: &crate::params::SynthParams,
) {
    let hz = midi_note_to_hz(snap.note) * libm::powf(2.0_f32, snap.pitch_bend / 12.0);
    vco.freq_hz = hz;
    vco.waveform = snap.waveform;
    filt.cutoff_hz = snap.cutoff_hz;
    filt.resonance = snap.resonance;
    vca.gain = snap.volume;

    if snap.gate {
        adsr.gate_on();
    } else {
        adsr.gate_off();
    }

    let mut audio: [f32; BLOCK_FRAMES] = [0.0; BLOCK_FRAMES];
    let mut env: [f32; BLOCK_FRAMES] = [0.0; BLOCK_FRAMES];

    vco.process(&mut audio);
    adsr.process(&mut env);
    filt.process(&mut audio);
    vca.apply_envelope(&mut audio, &env);

    for (i, &s) in audio.iter().enumerate() {
        let word = f32_to_i2s(s);
        buf[i * 2] = word; // Left
        buf[i * 2 + 1] = word; // Right (mono → stereo)
    }
}

#[task]
pub async fn audio_task(
    pio0: Peri<'static, PIO0>,
    bck: Peri<'static, PIN_9>,
    lrck: Peri<'static, PIN_10>,
    data: Peri<'static, PIN_11>,
    dma: Peri<'static, DMA_CH0>,
) {
    let Pio { mut common, sm0, .. } = Pio::new(pio0, Irqs);

    let program = PioI2sOutProgram::new(&mut common);
    let mut i2s = PioI2sOut::new(
        &mut common,
        sm0,
        dma,
        Irqs,
        data,
        bck,
        lrck,
        SAMPLE_RATE,
        BIT_DEPTH,
        &program,
    );
    i2s.start();

    defmt::info!("I2S PIO running: {}Hz, {}-bit", SAMPLE_RATE, BIT_DEPTH);

    // Initialise DSP modules.
    let init = { PARAMS.lock().await.clone() };
    let sr = SR_48000;

    let mut vco = Vco::new(sr, midi_note_to_hz(init.note), init.waveform);
    let mut adsr = Adsr::new(sr);
    let mut filt = Svf::new(sr, init.cutoff_hz, init.resonance, FilterMode::LowPass);
    let mut vca = Vca::new(init.volume);

    // SAFETY: we are the sole owner of these statics from this point on.
    let buf_a = unsafe { &mut *core::ptr::addr_of_mut!(DMA_BUF_A) };
    let buf_b = unsafe { &mut *core::ptr::addr_of_mut!(DMA_BUF_B) };

    // Pre-fill both buffers before starting the DMA loop.
    {
        let snap = PARAMS.lock().await.clone();
        render(buf_a, &mut vco, &mut adsr, &mut filt, &mut vca, &snap);
        render(buf_b, &mut vco, &mut adsr, &mut filt, &mut vca, &snap);
    }

    // Double-buffered audio loop:
    //   1. Send the active buffer via DMA.
    //   2. Render the next block into the inactive buffer.
    //   3. Await DMA completion and swap roles.
    loop {
        i2s.write(buf_a).await;
        {
            let snap = PARAMS.lock().await.clone();
            render(buf_b, &mut vco, &mut adsr, &mut filt, &mut vca, &snap);
        }

        i2s.write(buf_b).await;
        {
            let snap = PARAMS.lock().await.clone();
            render(buf_a, &mut vco, &mut adsr, &mut filt, &mut vca, &snap);
        }
    }
}