Rework and optimize block smoothing API
You now need to bring your own buffer instead of the smoother having a built in vector you would need to pre-allocate. This makes the API simpler, and also much more flexible when doing polyphonic modulation. In addition, the new API is much more efficient when there is no smoothing going on anymore.
This commit is contained in:
parent
89b2d0a66c
commit
68cf0455ee
|
@ -8,6 +8,12 @@ code then it will not be listed here.
|
|||
|
||||
## [2022-07-06]
|
||||
|
||||
- The block smoothing API has been reworked. Instead of `Smoother`s having their
|
||||
own built in block buffer, you now need to provide your own mutable slice for
|
||||
the smoother to fill. This makes the API easier to understand, more flexible,
|
||||
and it allows cloning smoothers without worrying about allocations for use
|
||||
with polyphonic modulation. In addition, the new implementation is much more
|
||||
efficient when the smoothing period has ended before or during the block.
|
||||
- There are new `NoteEvent::PolyModulation` and `NoteEvent::MonoAutomation` as
|
||||
part of polyphonic modulation support for CLAP plugins.
|
||||
|
||||
|
|
|
@ -83,10 +83,7 @@ impl<'a> Buffer<'a> {
|
|||
/// SIMD.
|
||||
///
|
||||
/// The parameter smoothers can also produce smoothed values for an entire block using
|
||||
/// [`Smoother::next_block()`][crate::prelude::Smoother::next_block()]. Before using this, you
|
||||
/// will need to call
|
||||
/// [`Plugin::initialize_block_smoothers()`][crate::prelude::Plugin::initialize_block_smoothers()]
|
||||
/// with the same `max_block_size` in your initialization function first.
|
||||
/// [`Smoother::next_block()`][crate::prelude::Smoother::next_block()].
|
||||
///
|
||||
/// You can use this to obtain block-slices from a buffer so you can pass them to a library:
|
||||
///
|
||||
|
|
|
@ -75,9 +75,7 @@
|
|||
//! plugins](https://github.com/robbert-vdh/nih-plug/tree/master/plugins).
|
||||
//! - After calling `.with_smoother()` during an integer or floating point parameter's creation,
|
||||
//! you can use `param.smoothed` to access smoothed values for that parameter. Be sure to check
|
||||
//! out the [`Smoother`][prelude::Smoother] API for more details. If you want to generate entire
|
||||
//! blocks of smoothed values, be sure to call the predefined
|
||||
//! `[Plugin::initialize_block_smoothers()]` method from your plugin's `initialize()` function.
|
||||
//! out the [`Smoother`][prelude::Smoother] API for more details.
|
||||
//!
|
||||
//! There's a whole lot more to discuss, but once you understand the above you should be able to
|
||||
//! figure out the rest by reading through the examples and the API documetnation. Good luck!
|
||||
|
|
|
@ -147,11 +147,6 @@ pub trait Param: Display {
|
|||
/// wrappers. This **does** snap to step sizes for continuous parameters (i.e. [`FloatParam`]).
|
||||
fn preview_plain(&self, normalized: f32) -> Self::Plain;
|
||||
|
||||
/// Allocate memory for block-based smoothing. The
|
||||
/// [`Plugin::initialize_block_smoothers()`][crate::prelude::Plugin::initialize_block_smoothers()] method
|
||||
/// will do this for every smoother.
|
||||
fn initialize_block_smoother(&mut self, max_block_size: usize);
|
||||
|
||||
/// Flags to control the parameter's behavior. See [`ParamFlags`].
|
||||
fn flags(&self) -> ParamFlags;
|
||||
|
||||
|
|
|
@ -143,8 +143,6 @@ impl Param for BoolParam {
|
|||
normalized > 0.5
|
||||
}
|
||||
|
||||
fn initialize_block_smoother(&mut self, _max_block_size: usize) {}
|
||||
|
||||
fn flags(&self) -> ParamFlags {
|
||||
self.flags
|
||||
}
|
||||
|
|
|
@ -175,10 +175,6 @@ impl<T: Enum + PartialEq> Param for EnumParam<T> {
|
|||
T::from_index(self.inner.preview_plain(normalized) as usize)
|
||||
}
|
||||
|
||||
fn initialize_block_smoother(&mut self, max_block_size: usize) {
|
||||
self.inner.initialize_block_smoother(max_block_size)
|
||||
}
|
||||
|
||||
fn flags(&self) -> ParamFlags {
|
||||
self.inner.flags()
|
||||
}
|
||||
|
@ -261,10 +257,6 @@ impl Param for EnumParamInner {
|
|||
self.inner.preview_plain(normalized)
|
||||
}
|
||||
|
||||
fn initialize_block_smoother(&mut self, max_block_size: usize) {
|
||||
self.inner.initialize_block_smoother(max_block_size)
|
||||
}
|
||||
|
||||
fn flags(&self) -> ParamFlags {
|
||||
self.inner.flags()
|
||||
}
|
||||
|
|
|
@ -201,10 +201,6 @@ impl Param for FloatParam {
|
|||
}
|
||||
}
|
||||
|
||||
fn initialize_block_smoother(&mut self, max_block_size: usize) {
|
||||
self.smoothed.initialize_block_smoother(max_block_size);
|
||||
}
|
||||
|
||||
fn flags(&self) -> ParamFlags {
|
||||
self.flags
|
||||
}
|
||||
|
|
|
@ -166,10 +166,6 @@ impl Param for IntParam {
|
|||
self.range.unnormalize(normalized)
|
||||
}
|
||||
|
||||
fn initialize_block_smoother(&mut self, max_block_size: usize) {
|
||||
self.smoothed.initialize_block_smoother(max_block_size);
|
||||
}
|
||||
|
||||
fn flags(&self) -> ParamFlags {
|
||||
self.flags
|
||||
}
|
||||
|
|
|
@ -161,7 +161,6 @@ impl ParamPtr {
|
|||
param_ptr_forward!(pub unsafe fn step_count(&self) -> Option<usize>);
|
||||
param_ptr_forward!(pub unsafe fn previous_normalized_step(&self, from: f32) -> f32);
|
||||
param_ptr_forward!(pub unsafe fn next_normalized_step(&self, from: f32) -> f32);
|
||||
param_ptr_forward!(pub unsafe fn initialize_block_smoother(&mut self, max_block_size: usize));
|
||||
param_ptr_forward!(pub unsafe fn normalized_value_to_string(&self, normalized: f32, include_unit: bool) -> String);
|
||||
param_ptr_forward!(pub unsafe fn string_to_normalized_value(&self, string: &str) -> Option<f32>);
|
||||
param_ptr_forward!(pub unsafe fn flags(&self) -> ParamFlags);
|
||||
|
|
|
@ -1,11 +1,8 @@
|
|||
//! Utilities to handle smoothing parameter changes over time.
|
||||
|
||||
use atomic_float::AtomicF32;
|
||||
use atomic_refcell::{AtomicRefCell, AtomicRefMut};
|
||||
use std::sync::atomic::{AtomicI32, Ordering};
|
||||
|
||||
use crate::buffer::Block;
|
||||
|
||||
/// Controls if and how parameters gets smoothed.
|
||||
#[derive(Debug, Clone, Copy)]
|
||||
pub enum SmoothingStyle {
|
||||
|
@ -55,15 +52,10 @@ pub struct Smoother<T> {
|
|||
current: AtomicF32,
|
||||
/// The value we're smoothing towards
|
||||
target: T,
|
||||
|
||||
/// A dense buffer containing smoothed values for an entire block of audio. Useful when using
|
||||
/// [`Buffer::iter_blocks()`][crate::prelude::Buffer::iter_blocks()] to process small blocks of audio
|
||||
/// multiple times.
|
||||
block_values: AtomicRefCell<Vec<T>>,
|
||||
}
|
||||
|
||||
/// An iterator that continuously produces smoothed values. Can be used as an alternative to the
|
||||
/// built-in block-based smoothing API. Since the iterator itself is infinite, you can use
|
||||
/// block-based smoothing API. Since the iterator itself is infinite, you can use
|
||||
/// [`Smoother::is_smoothing()`] and [`Smoother::steps_left()`] to get information on the current
|
||||
/// smoothing status.
|
||||
pub struct SmootherIter<'a, T> {
|
||||
|
@ -85,8 +77,6 @@ impl<T: Smoothable> Default for Smoother<T> {
|
|||
step_size: Default::default(),
|
||||
current: AtomicF32::new(0.0),
|
||||
target: Default::default(),
|
||||
|
||||
block_values: AtomicRefCell::new(Vec::new()),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
@ -143,15 +133,6 @@ impl<T: Smoothable> Smoother<T> {
|
|||
SmootherIter { smoother: self }
|
||||
}
|
||||
|
||||
/// Allocate memory to store smoothed values for an entire block of audio. Call this in
|
||||
/// [`Plugin::initialize()`][crate::prelude::Plugin::initialize()] with the same max block size you are
|
||||
/// going to pass to [`Buffer::iter_blocks()`][crate::prelude::Buffer::iter_blocks()].
|
||||
pub fn initialize_block_smoother(&mut self, max_block_size: usize) {
|
||||
self.block_values
|
||||
.borrow_mut()
|
||||
.resize_with(max_block_size, || T::default());
|
||||
}
|
||||
|
||||
/// Reset the smoother the specified value.
|
||||
pub fn reset(&mut self, value: T) {
|
||||
self.target = value;
|
||||
|
@ -248,43 +229,44 @@ impl<T: Smoothable> Smoother<T> {
|
|||
T::from_f32(self.current.load(Ordering::Relaxed))
|
||||
}
|
||||
|
||||
/// Produce smoothed values for an entire block of audio. Used in conjunction with
|
||||
/// [`Buffer::iter_blocks()`][crate::prelude::Buffer::iter_blocks()]. Make sure to call
|
||||
/// [`Plugin::initialize_block_smoothers()`][crate::prelude::Plugin::initialize_block_smoothers()] with
|
||||
/// the same maximum buffer block size as the one passed to `iter_blocks()` in your
|
||||
/// [`Plugin::initialize()`][crate::prelude::Plugin::initialize()] function first to allocate memory for
|
||||
/// the block smoothing.
|
||||
///
|
||||
/// Returns a `None` value if the block length exceed's the allocated capacity.
|
||||
/// Produce smoothed values for an entire block of audio. This is useful when iterating the same
|
||||
/// block of audio multiple times. For instance when summing voices for a synthesizer.
|
||||
/// `block_values[..block_len]` will be filled with the smoothed values. This is simply a
|
||||
/// convenient function for [`next_block_exact()`][Self::next_block_exact()] when iterating over
|
||||
/// variable length blocks with a known maximum size.
|
||||
///
|
||||
/// # Panics
|
||||
///
|
||||
/// Panics if this function is called again while another block value slice is still alive.
|
||||
pub fn next_block(&self, block: &Block) -> Option<AtomicRefMut<[T]>> {
|
||||
self.next_block_mapped(block, |x| x)
|
||||
/// Panics if `block_len > block_values.len()`.
|
||||
pub fn next_block(&self, block_values: &mut [T], block_len: usize) {
|
||||
self.next_block_exact_mapped(&mut block_values[..block_len], |x| x)
|
||||
}
|
||||
|
||||
/// The same as [`next_block()`][Self::next_block()], but filling the entire slice.
|
||||
pub fn next_block_exact(&self, block_values: &mut [T]) {
|
||||
self.next_block_exact_mapped(block_values, |x| x)
|
||||
}
|
||||
|
||||
/// The same as [`next_block()`][Self::next_block()], but with a function applied to each
|
||||
/// produced value. Useful when applying modulation to a smoothed parameter.
|
||||
pub fn next_block_mapped(
|
||||
&self,
|
||||
block: &Block,
|
||||
mut f: impl FnMut(T) -> T,
|
||||
) -> Option<AtomicRefMut<[T]>> {
|
||||
let mut block_values = self.block_values.borrow_mut();
|
||||
if block_values.len() < block.len() {
|
||||
return None;
|
||||
}
|
||||
pub fn next_block_mapped(&self, block_values: &mut [T], block_len: usize, f: impl Fn(T) -> T) {
|
||||
self.next_block_exact_mapped(&mut block_values[..block_len], f)
|
||||
}
|
||||
|
||||
// TODO: As a small optimization we could split this up into two loops for the smoothed and
|
||||
// unsmoothed parts. Another worthwhile optimization would be to remember if the
|
||||
// buffer is already filled with the target value and [Self::is_smoothing()] is false.
|
||||
// In that case we wouldn't need to do anything ehre.
|
||||
(&mut block_values[..block.len()]).fill_with(|| f(self.next()));
|
||||
/// The same as [`next_block_exact()`][Self::next_block()], but with a function applied to each
|
||||
/// produced value. Useful when applying modulation to a smoothed parameter.
|
||||
pub fn next_block_exact_mapped(&self, block_values: &mut [T], f: impl Fn(T) -> T) {
|
||||
// `self.next()` will yield the current value if the parameter is no longer smoothing, but
|
||||
// it's a bit of a waste to continuesly call that if only the first couple or none of the
|
||||
// values in `block_values` would require smoothing and the rest don't. Instead, we'll just
|
||||
// smooth the values as necessary, and then reuse the target value for the rest of the
|
||||
// block.
|
||||
let num_smoothed_values = block_values
|
||||
.len()
|
||||
.min(self.steps_left.load(Ordering::Relaxed) as usize);
|
||||
|
||||
Some(AtomicRefMut::map(block_values, |values| {
|
||||
&mut values[..block.len()]
|
||||
}))
|
||||
block_values[..num_smoothed_values].fill_with(|| f(self.next()));
|
||||
block_values[num_smoothed_values..].fill(self.target);
|
||||
}
|
||||
}
|
||||
|
||||
|
|
|
@ -152,10 +152,7 @@ pub trait Plugin: Default + Send + Sync + 'static {
|
|||
/// per-sample SIMD or excessive branching. The parameter smoothers can also work in both modes:
|
||||
/// use [`Smoother::next()`][crate::prelude::Smoother::next()] for per-sample processing, and
|
||||
/// [`Smoother::next_block()`][crate::prelude::Smoother::next_block()] for block-based
|
||||
/// processing. In order to use block-based smoothing, you will need to call
|
||||
/// [`initialize_block_smoothers()`][Self::initialize_block_smoothers()] in your
|
||||
/// [`initialize()`][Self::initialize()] function first to reserve enough capacity in the
|
||||
/// smoothers.
|
||||
/// processing.
|
||||
///
|
||||
/// The `context` object contains context information as well as callbacks for working with note
|
||||
/// events. The [`AuxiliaryBuffers`] contain the plugin's sidechain input buffers and
|
||||
|
@ -179,17 +176,6 @@ pub trait Plugin: Default + Send + Sync + 'static {
|
|||
/// `initialize()` may be called more than once before `deactivate()` is called, for instance
|
||||
/// when restoring state while the plugin is still activate.
|
||||
fn deactivate(&mut self) {}
|
||||
|
||||
/// Convenience function provided to allocate memory for block-based smoothing for this plugin.
|
||||
/// Since this allocates memory, this should be called in [`initialize()`][Self::initialize()].
|
||||
/// If you are going to use [`Buffer::iter_blocks()`] and want to use parameter smoothing in
|
||||
/// those blocks, then call this function with the same maximum block size first before calling
|
||||
/// [`Smoother::next_block()`][crate::prelude::Smoother::next_block()].
|
||||
fn initialize_block_smoothers(&mut self, max_block_size: usize) {
|
||||
for (_, mut param, _) in self.params().param_map() {
|
||||
unsafe { param.initialize_block_smoother(max_block_size) };
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// Provides auxiliary metadata needed for a CLAP plugin.
|
||||
|
|
Loading…
Reference in a new issue