Parameterize FirCoefficients over the kernel size

plugins/crossover/src/crossover/fir.rs

@@ -200,7 +200,7 @@ impl FirCrossover {
                 for coef in fir_hp_coefs.0.iter_mut() {
                     *coef = -*coef;
                 }
-                fir_hp_coefs.0[FILTER_SIZE / 2] += f32x2::splat(1.0);
+                fir_hp_coefs.0[FILTER_SIZE / 2] += 1.0;
 
                 self.band_filters[num_bands - 1].coefficients = fir_hp_coefs;
             }

plugins/crossover/src/crossover/fir/filter.rs

@@ -26,10 +26,7 @@ use crate::crossover::iir::biquad::{Biquad, BiquadCoefficients};
 pub struct FirFilter {
     /// The coefficients for this filter. The filters for both channels should be equivalent, this
     /// just avoids broadcasts in the filter process.
-    ///
+    pub coefficients: FirCoefficients<FILTER_SIZE>,
-    /// TODO: Profile to see if storing this as f32x2 rather than f32s plus splatting makes any
-    ///       difference in performance at all
-    pub coefficients: FirCoefficients,
 
     /// A ring buffer storing the last `FILTER_SIZE - 1` samples. The capacity is `FILTER_SIZE`
     /// rounded up to the next power of two.
@@ -40,11 +37,11 @@ pub struct FirFilter {
     delay_buffer_next_idx: usize,
 }
 
-/// Coefficients for an FIR filter. This struct includes ways to design the filter. Parameterized
+/// Coefficients for a (linear-phase) FIR filter. This struct includes ways to design the filter.
-/// over `f32x2` only for the time being since that's what we need here.
+/// `T` is the sample type and `N` is the number of taps/coefficients and should be odd for linear-phase filters.
 #[repr(transparent)]
 #[derive(Debug, Clone)]
-pub struct FirCoefficients(pub [f32x2; FILTER_SIZE]);
+pub struct FirCoefficients<const N: usize>(pub [f32; N]);
 
 impl Default for FirFilter {
     fn default() -> Self {
@@ -56,12 +53,12 @@ impl Default for FirFilter {
     }
 }
 
-impl Default for FirCoefficients {
+impl<const N: usize> Default for FirCoefficients<N> {
     fn default() -> Self {
         // Initialize this to a delay with the same amount of latency as we'd introduce with our
         // linear-phase filters
-        let mut coefficients = [f32x2::default(); FILTER_SIZE];
+        let mut coefficients = [0.0; N];
-        coefficients[FILTER_SIZE / 2] = f32x2::splat(1.0);
+        coefficients[N / 2] = 1.0;
 
         Self(coefficients)
     }
@@ -73,7 +70,7 @@ impl FirFilter {
         // TODO: Replace direct convolution with FFT convolution, would make the implementation much
         //       more complex though because of the multi output part
         let coefficients = &self.coefficients.0;
-        let mut result = coefficients[0] * samples;
+        let mut result = f32x2::splat(coefficients[0]) * samples;
 
         // Now multiply `self.coefficients[1..]` with the delay buffer starting at
         // `self.delay_buffer_next_idx - 1`, wrapping around to the end when that is reached
@@ -89,7 +86,7 @@ impl FirFilter {
                 .rev(),
         ) {
             // `result += coefficient * sample`, but with explicit FMA
-            result = coefficient.mul_add(*delayed_sample, result);
+            result = f32x2::splat(*coefficient).mul_add(*delayed_sample, result);
         }
 
         let after_wraparound_begin_idx =
@@ -100,7 +97,7 @@ impl FirFilter {
                 .iter()
                 .rev(),
         ) {
-            result = coefficient.mul_add(*delayed_sample, result);
+            result = f32x2::splat(*coefficient).mul_add(*delayed_sample, result);
         }
 
         // And finally write the samples to the delay buffer for the enxt sample
@@ -117,7 +114,7 @@ impl FirFilter {
     }
 }
 
-impl FirCoefficients {
+impl<const N: usize> FirCoefficients<N> {
     /// A somewhat crude but very functional and relatively fast way create linear phase FIR
     /// **low-pass** filter that matches the frequency response of a fourth order biquad low-pass
     /// filter. As in, this matches the frequency response magnitudes of applying those biquads to a
@@ -150,23 +147,24 @@ impl FirCoefficients {
     ///
     /// The corresponding high-pass filter can be computed through spectral inversion.
     pub fn design_fourth_order_linear_phase_low_pass_from_biquad(
-        biquad_coefs: BiquadCoefficients<f32x2>,
+        biquad_coefs: BiquadCoefficients<f32>,
     ) -> Self {
-        const CENTER_IDX: usize = FILTER_SIZE / 2;
+        // Ruest doesn't allow you to define this as a constant
+        let center_idx = N / 2;
 
         // We'll start with an impulse (at exactly half of this odd sized buffer)...
-        let mut impulse_response = [f32x2::default(); FILTER_SIZE];
+        let mut impulse_response = [0.0; N];
-        impulse_response[CENTER_IDX] = f32x2::splat(1.0);
+        impulse_response[center_idx] = 1.0;
 
         // ...and filter that in both directions
         let mut biquad = Biquad::default();
         biquad.coefficients = biquad_coefs;
-        for sample in impulse_response.iter_mut().skip(CENTER_IDX - 1) {
+        for sample in impulse_response.iter_mut().skip(center_idx - 1) {
             *sample = biquad.process(*sample);
         }
 
         biquad.reset();
-        for sample in impulse_response.iter_mut().skip(CENTER_IDX - 1).rev() {
+        for sample in impulse_response.iter_mut().skip(center_idx - 1).rev() {
             *sample = biquad.process(*sample);
         }
 
@@ -176,19 +174,19 @@ impl FirCoefficients {
 
         // Adopted from `nih_plug::util::window`. We only end up applying the right half of the
         // window, starting at the top of the window.
-        let blackman_scale_1 = (2.0 * f32::consts::PI) / (impulse_response.len() - 1) as f32;
+        let blackman_scale_1 = (2.0 * f32::consts::PI) / (N - 1) as f32;
         let blackman_scale_2 = blackman_scale_1 * 2.0;
-        for (sample_idx, sample) in impulse_response.iter_mut().enumerate().skip(CENTER_IDX - 1) {
+        for (sample_idx, sample) in impulse_response.iter_mut().enumerate().skip(center_idx - 1) {
             let cos_1 = (blackman_scale_1 * sample_idx as f32).cos();
             let cos_2 = (blackman_scale_2 * sample_idx as f32).cos();
-            *sample *= f32x2::splat(0.42 - (0.5 * cos_1) + (0.08 * cos_2));
+            *sample *= 0.42 - (0.5 * cos_1) + (0.08 * cos_2);
         }
 
         // Since this final filter will be symmetrical around `impulse_response[CENTER_IDX]`, we
         // can simply normalize based on that fact:
-        let would_be_impulse_response_sum =
+        let would_be_impulse_response_sum = (impulse_response.iter().skip(center_idx).sum::<f32>()
-            (impulse_response.iter().skip(CENTER_IDX).sum::<f32x2>() * f32x2::splat(2.0))
+            * 2.0)
-                - impulse_response[CENTER_IDX];
+            - impulse_response[center_idx];
         let would_be_impulse_response_recip = would_be_impulse_response_sum.recip();
         for sample in &mut impulse_response {
             *sample *= would_be_impulse_response_recip;
@@ -196,8 +194,8 @@ impl FirCoefficients {
 
         // And finally we can simply copy the right half of the filter kernel to the left half
         // around the `CENTER_IDX`.
-        for source_idx in CENTER_IDX + 1..impulse_response.len() {
+        for source_idx in center_idx + 1..N {
-            let target_idx = CENTER_IDX - (source_idx - CENTER_IDX);
+            let target_idx = center_idx - (source_idx - center_idx);
             impulse_response[target_idx] = impulse_response[source_idx];
         }