Struct BlipBuf

Source

pub struct BlipBuf { /* private fields */ }

Expand description

Streaming BLEP decimator + filter chain that feeds host-rate samples to the frontend’s audio thread.

Implementations§

Source §

impl BlipBuf

Source

pub fn new(sample_rate: u32, cpu_rate: f64) -> Self

Create a new band-limited buffer.

sample_rate is the host audio rate in Hz (typically 44 100). cpu_rate is the NES CPU rate in Hz (CPU_HZ_NTSC or CPU_HZ_PAL).

Source

pub fn set_filter_model(&mut self, model: FilterModel)

v2.1.3 — swap the analog output-filter model (see crate::mixer::FilterModel), rebuilding the filter chain at the current sample rate. This resets the filter’s IIR state, so switching while audio is playing (the Settings selector applies it live) produces a brief transient — as with any live filter swap. The frontend also applies it at ROM load / power-cycle, where there is no audible discontinuity.

Source

pub fn reset(&mut self)

Reset to silence. Empties the input ring, the pending output queue, and the filter chain state.

Source

pub fn add_sample(&mut self, value: f32)

Add one mixed sample at CPU resolution. The buffer accumulates host-rate samples internally; drain via Self::drain or Self::drain_all.

The mixer’s output is approximately in [-0.5, 0.5] for the 2A03 alone, with on-cart audio expansions pushing the absolute range up somewhat. We clamp value defensively so a stray NaN/Inf or runaway mapper can’t propagate non-finite values into the FIR state. The clamp range is far outside any sane mixer output, so the gate is purely a saturation backstop, not a normal-operation limiter.

Source

pub fn drain(&mut self, out: &mut [f32]) -> usize

Drain finalized samples into out. Returns the number written. Excess pending samples are kept; if out.len() < self.samples.len() the remainder waits for the next drain.

Source

pub fn drain_all(&mut self) -> Vec<f32>

Drain all finalized samples into a new Vec.

The filled buffer goes to the caller, and a fresh one with the same capacity takes its place (v2.7.5, core audit IMP-07). A buffer returned by value must be replaced every call whatever happens; the choice is only whether its replacement starts empty. With mem::take it started at capacity zero and regrew by doubling through the next frame’s pushes (about nine reallocations for a frame’s ~800 samples); now it is one allocation of the size the last frame needed, up to DRAIN_ALL_KEEP_CAPACITY samples. Measured -0.89% frame time on a steady-state frame-then-drain workload, reproduced on two runs (docs/performance.md §v2.7.5). Callers that drain into their own buffer with Self::drain allocate nothing and are unaffected. The samples themselves are untouched, so the output is byte-identical.

Source