pxdct 0.2.4

/*
 * // Copyright (c) Radzivon Bartoshyk 01/2026. All rights reserved.
 * //
 * // Redistribution and use in source and binary forms, with or without modification,
 * // are permitted provided that the following conditions are met:
 * //
 * // 1.  Redistributions of source code must retain the above copyright notice, this
 * // list of conditions and the following disclaimer.
 * //
 * // 2.  Redistributions in binary form must reproduce the above copyright notice,
 * // this list of conditions and the following disclaimer in the documentation
 * // and/or other materials provided with the distribution.
 * //
 * // 3.  Neither the name of the copyright holder nor the names of its
 * // contributors may be used to endorse or promote products derived from
 * // this software without specific prior written permission.
 * //
 * // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
 * // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
 * // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
 * // DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
 * // FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
 * // DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
 * // SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
 * // CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
 * // OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * // OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */
use crate::mla::fmla;
use crate::neon::dct4::radix2f::dct4_radix2f_rotation_twiddles_neon;
use crate::neon::util::NeonStoreF;
use crate::util::try_vec;
use crate::{PxdctError, PxdctExecutor};
use num_traits::One;
use std::sync::Arc;

pub(crate) struct NeonDct4MixedRadix2f {
    dct2: Arc<dyn PxdctExecutor<f32> + Send + Sync>,
    twiddles: Vec<NeonStoreF>,
    execution_length: usize,
}

impl NeonDct4MixedRadix2f {
    pub(crate) fn new(
        len: usize,
        dct2: Arc<dyn PxdctExecutor<f32> + Send + Sync>,
    ) -> Result<Self, PxdctError> {
        assert_eq!(
            dct2.length(),
            len / 2,
            "DCT-II even length DCTs must be half of DCT-IV"
        );
        let inner_len = dct2.length();

        Ok(Self {
            dct2,
            twiddles: dct4_radix2f_rotation_twiddles_neon(inner_len, len),
            execution_length: len,
        })
    }
}

impl PxdctExecutor<f32> for NeonDct4MixedRadix2f {
    fn execute(&self, data: &mut [f32]) -> Result<(), PxdctError> {
        if !data.len().is_multiple_of(self.execution_length) {
            return Err(PxdctError::InvalidSizeMultiplier(
                data.len(),
                self.execution_length,
            ));
        }
        let mut scratch = try_vec![f32::default(); self.execution_length];

        let len = self.length();
        let half_len = len / 2;

        // This kernel implements a radix-2 DCT-IV using one inner DCT-II.
        // It follows the classical even/odd pre-rotation → DCT-II → post-butterfly scheme.
        //
        // For each length-N block:
        //
        //   DCT4(N)  =  PostRotate · ( DCT2(N/2) ⊕ DCT2(N/2) ) · PreRotate
        //
        // where the pre/post rotations are fused with alternating sign flips to minimize
        // twiddle storage and multiplications.
        for chunk in data.chunks_exact_mut(self.execution_length) {
            let (left, right) = scratch.split_at_mut(half_len);

            let mut k = 0usize;
            let mut tk = 0usize;
            let signs_re = NeonStoreF::set_values(-0.0, 0.0, -0.0, 0.0);
            let signs_im = NeonStoreF::set_values(0.0, -0.0, 0.0, -0.0);

            let inner_len = self.dct2.length();

            // -------- Pre-rotation / even-odd folding --------
            // Fold symmetric samples (x[i], x[N-1-i]) into two half-length sequences.
            while k + 4 <= inner_len {
                const S: usize = 4;
                let front = NeonStoreF::load(unsafe { chunk.get_unchecked(k..) });
                let back =
                    NeonStoreF::load(unsafe { chunk.get_unchecked(len - k - S..) }).reverse();

                let twiddle_re = unsafe { *self.twiddles.get_unchecked(tk) };
                let twiddle_im = unsafe { *self.twiddles.get_unchecked(tk + 1) };

                let ll = fmla(twiddle_re, front, twiddle_im * back);
                unsafe {
                    ll.write(left.get_unchecked_mut(k..));
                }
                let rr = fmla(
                    twiddle_re.xor(signs_re),
                    back,
                    twiddle_im.xor(signs_im) * front,
                );
                unsafe {
                    rr.reverse()
                        .write(right.get_unchecked_mut(half_len - k - S..));
                }

                tk += 2;
                k += 4;
            }

            let rem = inner_len - k;
            if rem == 3 {
                const S: usize = 3;
                let front = NeonStoreF::load3(unsafe { chunk.get_unchecked(k..) });
                let back =
                    NeonStoreF::load3(unsafe { chunk.get_unchecked(len - k - S..) }).reverse3();

                let twiddle_re = unsafe { *self.twiddles.get_unchecked(tk) };
                let twiddle_im = unsafe { *self.twiddles.get_unchecked(tk + 1) };

                let ll = fmla(twiddle_re, front, twiddle_im * back);
                unsafe {
                    ll.write3(left.get_unchecked_mut(k..));
                }
                let rr = fmla(
                    twiddle_re.xor(signs_re),
                    back,
                    twiddle_im.xor(signs_im) * front,
                );
                unsafe {
                    rr.reverse3()
                        .write3(right.get_unchecked_mut(half_len - k - S..));
                }
            } else if rem == 2 {
                const S: usize = 2;
                let front = NeonStoreF::load2(unsafe { chunk.get_unchecked(k..) });
                let back =
                    NeonStoreF::load2(unsafe { chunk.get_unchecked(len - k - S..) }).reverse2();

                let twiddle_re = unsafe { *self.twiddles.get_unchecked(tk) };
                let twiddle_im = unsafe { *self.twiddles.get_unchecked(tk + 1) };

                let ll = fmla(twiddle_re, front, twiddle_im * back);
                unsafe {
                    ll.write2(left.get_unchecked_mut(k..));
                }
                let rr = fmla(
                    twiddle_re.xor(signs_re),
                    back,
                    twiddle_im.xor(signs_im) * front,
                );
                unsafe {
                    rr.reverse2()
                        .write2(right.get_unchecked_mut(half_len - k - S..));
                }
            } else if rem == 1 {
                const S: usize = 1;
                let front = NeonStoreF::load1(unsafe { chunk.get_unchecked(k..) });
                let back = NeonStoreF::load1(unsafe { chunk.get_unchecked(len - k - S..) });

                let twiddle_re = unsafe { *self.twiddles.get_unchecked(tk) };
                let twiddle_im = unsafe { *self.twiddles.get_unchecked(tk + 1) };

                let ll = fmla(twiddle_re, front, twiddle_im * back);
                unsafe {
                    ll.write1(left.get_unchecked_mut(k..));
                }
                let rr = fmla(
                    twiddle_re.xor(signs_re),
                    back,
                    twiddle_im.xor(signs_im) * front,
                );
                unsafe {
                    rr.write1(right.get_unchecked_mut(half_len - k - S..));
                }
            }

            self.dct2.execute(&mut scratch)?;

            let (left, right) = scratch.split_at_mut(half_len);

            chunk[0] = left[0];
            chunk[len - 1] = right[0];

            let signs_re = if half_len.is_multiple_of(2) {
                NeonStoreF::set_values(-1.0, 1.0, -1.0, 1.0)
            } else {
                NeonStoreF::set_values(1.0, -1.0, 1.0, -1.0)
            };
            let signs_im = if half_len.is_multiple_of(2) {
                NeonStoreF::set_values(1.0, -1.0, 1.0, -1.0)
            } else {
                NeonStoreF::set_values(-1.0, 1.0, -1.0, 1.0)
            };

            let mut i = 1usize;
            while i + 4 < half_len {
                let il = NeonStoreF::load(unsafe { left.get_unchecked(i..) });
                let rr =
                    NeonStoreF::load(unsafe { right.get_unchecked(half_len - i - 3..) }).reverse();

                unsafe {
                    let q = i - 1;
                    let u0 = fmla(signs_re, rr, il);
                    let u1 = fmla(signs_im, rr, il);
                    let interleaved = u0.zip(u1);
                    interleaved[0].write(chunk.get_unchecked_mut(q * 2 + 1..));
                    interleaved[1].write(chunk.get_unchecked_mut(q * 2 + 5..));
                }

                i += 4;
            }

            let mut sign_left = if half_len.is_multiple_of(2) {
                -f32::one()
            } else {
                f32::one()
            };
            let mut sign_right = if half_len.is_multiple_of(2) {
                f32::one()
            } else {
                -f32::one()
            };

            // -------- Post-butterfly recombination --------
            // Interleave even and odd spectra back into full DCT-IV ordering.
            for i in 1..half_len {
                let il = unsafe { *left.get_unchecked(i) };
                let rr = unsafe { *right.get_unchecked(half_len - i) };

                unsafe {
                    let q = i - 1;
                    *chunk.get_unchecked_mut(q * 2 + 1) = fmla(sign_left, rr, il);
                    *chunk.get_unchecked_mut(q * 2 + 2) = fmla(sign_right, rr, il);
                }

                sign_left = -sign_left;
                sign_right = -sign_right;
            }
        }

        Ok(())
    }

    #[inline]
    fn length(&self) -> usize {
        self.execution_length
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::dct2::prime_butterflies::Dct2Butterfly17;
    use crate::tests::naive_dct4_f32;
    use rand::Rng;

    #[test]
    fn test_split_dct4() {
        let mut input = vec![0.; 34];
        for z in input.iter_mut() {
            *z = rand::rng().random_range(1.0..2.0);
        }
        let reference_input = input.clone();
        let reference_input = naive_dct4_f32(&reference_input);
        let bf = NeonDct4MixedRadix2f::new(34, Arc::new(Dct2Butterfly17::default())).unwrap();
        bf.execute(&mut input).unwrap();
        input
            .iter()
            .zip(reference_input.iter())
            .enumerate()
            .for_each(|(i, (&src, &r0))| {
                assert!(
                    (src - r0).abs() < 1e-3,
                    "Difference must be < {}, but it was {}, at position {i}",
                    1e-3,
                    (src - r0).abs()
                )
            });
    }
}