vector_ta/indicators/

acosc.rs

use crate::utilities::data_loader::Candles;
use crate::utilities::enums::Kernel;
use crate::utilities::helpers::{
    alloc_with_nan_prefix, detect_best_batch_kernel, detect_best_kernel, init_matrix_prefixes,
    make_uninit_matrix,
};
#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
use core::arch::x86_64::*;
use std::mem::ManuallyDrop;
use thiserror::Error;

#[derive(Debug, Clone)]
pub enum AcoscData<'a> {
    Candles { candles: &'a Candles },
    Slices { high: &'a [f64], low: &'a [f64] },
}

#[derive(Debug, Clone, Default)]
pub struct AcoscParams {}

#[derive(Debug, Clone)]
pub struct AcoscInput<'a> {
    pub data: AcoscData<'a>,
    pub params: AcoscParams,
}
impl<'a> AcoscInput<'a> {
    #[inline]
    pub fn from_candles(candles: &'a Candles, params: AcoscParams) -> Self {
        Self {
            data: AcoscData::Candles { candles },
            params,
        }
    }
    #[inline]
    pub fn from_slices(high: &'a [f64], low: &'a [f64], params: AcoscParams) -> Self {
        Self {
            data: AcoscData::Slices { high, low },
            params,
        }
    }
    #[inline]
    pub fn with_default_candles(candles: &'a Candles) -> Self {
        Self {
            data: AcoscData::Candles { candles },
            params: AcoscParams::default(),
        }
    }
}

#[derive(Debug, Clone)]
pub struct AcoscOutput {
    pub osc: Vec<f64>,
    pub change: Vec<f64>,
}

#[derive(Debug, Error)]
pub enum AcoscError {
    #[error("acosc: Failed to get high/low fields from candles: {msg}")]
    CandleFieldError { msg: String },
    #[error(
        "acosc: Mismatch in high/low candle data lengths: high_len={high_len}, low_len={low_len}"
    )]
    LengthMismatch { high_len: usize, low_len: usize },
    #[error("acosc: Empty input data")]
    EmptyInputData,
    #[error("acosc: Not enough data: all values are NaN")]
    AllValuesNaN,
    #[error("acosc: Invalid period: period={period}, data_len={data_len}")]
    InvalidPeriod { period: usize, data_len: usize },
    #[error("acosc: Not enough data: needed={needed}, valid={valid}")]
    NotEnoughValidData { needed: usize, valid: usize },
    #[error("acosc: Output length mismatch: expected={expected}, got={got}")]
    OutputLengthMismatch { expected: usize, got: usize },
    #[error("acosc: Invalid range: start={start}, end={end}, step={step}")]
    InvalidRange { start: i64, end: i64, step: i64 },
    #[error("acosc: Invalid kernel for batch operation. Expected batch kernel, got: {0:?}")]
    InvalidKernelForBatch(Kernel),

    #[error("acosc: Not enough data points: required={required}, actual={actual}")]
    NotEnoughData { required: usize, actual: usize },
    #[error("acosc: Invalid kernel for batch operation. Expected batch kernel, got: {kernel:?}")]
    InvalidBatchKernel { kernel: Kernel },
}

#[inline]
pub fn acosc(input: &AcoscInput) -> Result<AcoscOutput, AcoscError> {
    acosc_with_kernel(input, Kernel::Auto)
}

#[inline(always)]
fn acosc_prepare<'a>(
    input: &'a AcoscInput,
    kernel: Kernel,
) -> Result<(&'a [f64], &'a [f64], usize, Kernel), AcoscError> {
    let (high, low) = match &input.data {
        AcoscData::Candles { candles } => {
            let h = candles
                .select_candle_field("high")
                .map_err(|e| AcoscError::CandleFieldError { msg: e.to_string() })?;
            let l = candles
                .select_candle_field("low")
                .map_err(|e| AcoscError::CandleFieldError { msg: e.to_string() })?;
            (h, l)
        }
        AcoscData::Slices { high, low } => (*high, *low),
    };

    if high.len() != low.len() {
        return Err(AcoscError::LengthMismatch {
            high_len: high.len(),
            low_len: low.len(),
        });
    }

    let len = high.len();
    if len == 0 {
        return Err(AcoscError::EmptyInputData);
    }
    const REQUIRED_LENGTH: usize = 39;

    let first = (0..len)
        .find(|&i| !high[i].is_nan() && !low[i].is_nan())
        .unwrap_or(len);
    let valid = len.saturating_sub(first);
    if valid == 0 {
        return Err(AcoscError::AllValuesNaN);
    }
    if valid < REQUIRED_LENGTH {
        return Err(AcoscError::NotEnoughValidData {
            needed: REQUIRED_LENGTH,
            valid,
        });
    }

    let chosen = match kernel {
        Kernel::Auto => detect_best_kernel(),
        other => other,
    };
    Ok((high, low, first, chosen))
}
pub fn acosc_with_kernel(input: &AcoscInput, kernel: Kernel) -> Result<AcoscOutput, AcoscError> {
    let (high, low, first, chosen) = acosc_prepare(input, kernel)?;

    let len = low.len();
    const WARMUP: usize = 38;
    let warmup_end = first + WARMUP;

    let mut osc = alloc_with_nan_prefix(len, warmup_end);
    let mut change = alloc_with_nan_prefix(len, warmup_end);

    if first < len {
        let valid_len = len - first;
        if valid_len > WARMUP {
            acosc_compute_into(
                &high[first..],
                &low[first..],
                chosen,
                &mut osc[first..],
                &mut change[first..],
            );
        }
    }

    Ok(AcoscOutput { osc, change })
}

#[inline(always)]
fn acosc_compute_into(
    high: &[f64],
    low: &[f64],
    kernel: Kernel,
    osc_out: &mut [f64],
    change_out: &mut [f64],
) {
    unsafe {
        match kernel {
            Kernel::Scalar | Kernel::ScalarBatch => acosc_scalar(high, low, osc_out, change_out),
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx2 | Kernel::Avx2Batch => acosc_avx2(high, low, osc_out, change_out),
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx512 | Kernel::Avx512Batch => acosc_avx512(high, low, osc_out, change_out),
            #[cfg(not(all(feature = "nightly-avx", target_arch = "x86_64")))]
            Kernel::Avx2 | Kernel::Avx2Batch | Kernel::Avx512 | Kernel::Avx512Batch => {
                acosc_scalar(high, low, osc_out, change_out)
            }
            Kernel::Auto => {
                unreachable!("Kernel::Auto should be resolved before calling compute_into")
            }
        }
    }
}

#[inline(always)]
pub fn acosc_scalar(high: &[f64], low: &[f64], osc: &mut [f64], change: &mut [f64]) {
    const PERIOD_SMA5: usize = 5;
    const PERIOD_SMA34: usize = 34;
    const INV5: f64 = 1.0 / 5.0;
    const INV34: f64 = 1.0 / 34.0;
    let len = high.len();
    debug_assert_eq!(low.len(), len);
    debug_assert_eq!(osc.len(), len);
    debug_assert_eq!(change.len(), len);
    debug_assert!(len >= PERIOD_SMA34 + PERIOD_SMA5);
    let mut queue5 = [0.0; PERIOD_SMA5];
    let mut queue34 = [0.0; PERIOD_SMA34];
    let mut queue5_ao = [0.0; PERIOD_SMA5];
    let mut sum5 = 0.0;
    let mut sum34 = 0.0;
    let mut sum5_ao = 0.0;
    let mut idx5 = 0;
    let mut idx34 = 0;
    let mut idx5_ao = 0;

    unsafe {
        let h_ptr = high.as_ptr();
        let l_ptr = low.as_ptr();
        let osc_ptr = osc.as_mut_ptr();
        let ch_ptr = change.as_mut_ptr();

        for i in 0..PERIOD_SMA34 {
            let med = (*h_ptr.add(i) + *l_ptr.add(i)) * 0.5;
            sum34 += med;
            queue34[i] = med;
            if i < PERIOD_SMA5 {
                sum5 += med;
                queue5[i] = med;
            }
        }
        for i in PERIOD_SMA34..(PERIOD_SMA34 + PERIOD_SMA5 - 1) {
            let med = (*h_ptr.add(i) + *l_ptr.add(i)) * 0.5;
            sum34 += med - queue34[idx34];
            queue34[idx34] = med;
            idx34 += 1;
            if idx34 == PERIOD_SMA34 {
                idx34 = 0;
            }
            let sma34 = sum34 * INV34;
            sum5 += med - queue5[idx5];
            queue5[idx5] = med;
            idx5 += 1;
            if idx5 == PERIOD_SMA5 {
                idx5 = 0;
            }
            let sma5 = sum5 * INV5;
            let ao = sma5 - sma34;
            sum5_ao += ao;
            queue5_ao[idx5_ao] = ao;
            idx5_ao += 1;
        }
        if idx5_ao == PERIOD_SMA5 {
            idx5_ao = 0;
        }
        let mut prev_res = 0.0;
        for i in (PERIOD_SMA34 + PERIOD_SMA5 - 1)..len {
            let med = (*h_ptr.add(i) + *l_ptr.add(i)) * 0.5;
            sum34 += med - queue34[idx34];
            queue34[idx34] = med;
            idx34 += 1;
            if idx34 == PERIOD_SMA34 {
                idx34 = 0;
            }
            let sma34 = sum34 * INV34;
            sum5 += med - queue5[idx5];
            queue5[idx5] = med;
            idx5 += 1;
            if idx5 == PERIOD_SMA5 {
                idx5 = 0;
            }
            let sma5 = sum5 * INV5;
            let ao = sma5 - sma34;
            let old_ao = queue5_ao[idx5_ao];
            sum5_ao += ao - old_ao;
            queue5_ao[idx5_ao] = ao;
            idx5_ao += 1;
            if idx5_ao == PERIOD_SMA5 {
                idx5_ao = 0;
            }
            let sma5_ao = sum5_ao * INV5;
            let res = ao - sma5_ao;
            let mom = res - prev_res;
            prev_res = res;
            *osc_ptr.add(i) = res;
            *ch_ptr.add(i) = mom;
        }
    }
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline]
pub fn acosc_avx512(high: &[f64], low: &[f64], osc: &mut [f64], change: &mut [f64]) {
    acosc_scalar(high, low, osc, change)
}
#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline]
pub fn acosc_avx2(high: &[f64], low: &[f64], osc: &mut [f64], change: &mut [f64]) {
    acosc_scalar(high, low, osc, change)
}
#[inline]
pub fn acosc_avx512_short(high: &[f64], low: &[f64], osc: &mut [f64], change: &mut [f64]) {
    acosc_scalar(high, low, osc, change)
}
#[inline]
pub fn acosc_avx512_long(high: &[f64], low: &[f64], osc: &mut [f64], change: &mut [f64]) {
    acosc_scalar(high, low, osc, change)
}

#[derive(Debug, Clone)]
pub struct AcoscStream {
    queue5: [f64; 5],
    queue34: [f64; 34],
    queue5_ao: [f64; 5],
    sum5: f64,
    sum34: f64,
    sum5_ao: f64,
    idx5: usize,
    idx34: usize,
    idx5_ao: usize,
    filled: usize,
    prev_res: f64,
}
impl AcoscStream {
    pub fn try_new(_params: AcoscParams) -> Result<Self, AcoscError> {
        Ok(Self {
            queue5: [0.0; 5],
            queue34: [0.0; 34],
            queue5_ao: [0.0; 5],
            sum5: 0.0,
            sum34: 0.0,
            sum5_ao: 0.0,
            idx5: 0,
            idx34: 0,
            idx5_ao: 0,
            filled: 0,
            prev_res: 0.0,
        })
    }
    #[inline(always)]
    pub fn update(&mut self, high: f64, low: f64) -> Option<(f64, f64)> {
        const PERIOD_SMA5: usize = 5;
        const PERIOD_SMA34: usize = 34;
        const INV5: f64 = 1.0 / 5.0;
        const INV34: f64 = 1.0 / 34.0;

        let med = (high + low) * 0.5;

        self.filled += 1;

        if self.filled <= PERIOD_SMA34 {
            self.sum34 += med;
            self.queue34[self.filled - 1] = med;

            if self.filled <= PERIOD_SMA5 {
                self.sum5 += med;
                self.queue5[self.filled - 1] = med;
            }
            return None;
        }

        if self.filled < (PERIOD_SMA34 + PERIOD_SMA5) {
            let old34 = self.queue34[self.idx34];
            self.sum34 += med - old34;
            self.queue34[self.idx34] = med;
            self.idx34 += 1;
            if self.idx34 == PERIOD_SMA34 {
                self.idx34 = 0;
            }
            let sma34 = self.sum34 * INV34;

            let old5 = self.queue5[self.idx5];
            self.sum5 += med - old5;
            self.queue5[self.idx5] = med;
            self.idx5 += 1;
            if self.idx5 == PERIOD_SMA5 {
                self.idx5 = 0;
            }
            let sma5 = self.sum5 * INV5;

            let ao = sma5 - sma34;
            self.sum5_ao += ao;
            self.queue5_ao[self.idx5_ao] = ao;
            self.idx5_ao += 1;
            if self.idx5_ao == PERIOD_SMA5 {
                self.idx5_ao = 0;
            }
            return None;
        }

        let old34 = self.queue34[self.idx34];
        self.sum34 += med - old34;
        self.queue34[self.idx34] = med;
        self.idx34 += 1;
        if self.idx34 == PERIOD_SMA34 {
            self.idx34 = 0;
        }
        let sma34 = self.sum34 * INV34;

        let old5 = self.queue5[self.idx5];
        self.sum5 += med - old5;
        self.queue5[self.idx5] = med;
        self.idx5 += 1;
        if self.idx5 == PERIOD_SMA5 {
            self.idx5 = 0;
        }
        let sma5 = self.sum5 * INV5;

        let ao = sma5 - sma34;
        let old_ao = self.queue5_ao[self.idx5_ao];
        self.sum5_ao += ao - old_ao;
        self.queue5_ao[self.idx5_ao] = ao;
        self.idx5_ao += 1;
        if self.idx5_ao == PERIOD_SMA5 {
            self.idx5_ao = 0;
        }

        let sma5_ao = self.sum5_ao * INV5;

        let res = ao - sma5_ao;
        let mom = res - self.prev_res;
        self.prev_res = res;

        Some((res, mom))
    }
}

#[derive(Clone, Debug)]
pub struct AcoscBatchRange {}

impl Default for AcoscBatchRange {
    fn default() -> Self {
        Self {}
    }
}

#[derive(Clone, Debug, Default)]
pub struct AcoscBatchBuilder {
    kernel: Kernel,
}
impl AcoscBatchBuilder {
    pub fn new() -> Self {
        Self::default()
    }
    pub fn kernel(mut self, k: Kernel) -> Self {
        self.kernel = k;
        self
    }
    pub fn apply_slice(self, high: &[f64], low: &[f64]) -> Result<AcoscBatchOutput, AcoscError> {
        acosc_batch_with_kernel(high, low, self.kernel)
    }
    pub fn with_default_slice(
        high: &[f64],
        low: &[f64],
        k: Kernel,
    ) -> Result<AcoscBatchOutput, AcoscError> {
        AcoscBatchBuilder::new().kernel(k).apply_slice(high, low)
    }
    pub fn apply_candles(self, c: &Candles) -> Result<AcoscBatchOutput, AcoscError> {
        let high = c
            .select_candle_field("high")
            .map_err(|e| AcoscError::CandleFieldError { msg: e.to_string() })?;
        let low = c
            .select_candle_field("low")
            .map_err(|e| AcoscError::CandleFieldError { msg: e.to_string() })?;
        self.apply_slice(high, low)
    }
    pub fn with_default_candles(c: &Candles) -> Result<AcoscBatchOutput, AcoscError> {
        AcoscBatchBuilder::new()
            .kernel(Kernel::Auto)
            .apply_candles(c)
    }
}
#[derive(Clone, Debug)]
pub struct AcoscBatchOutput {
    pub osc: Vec<f64>,
    pub change: Vec<f64>,
    pub rows: usize,
    pub cols: usize,
}
pub fn acosc_batch_with_kernel(
    high: &[f64],
    low: &[f64],
    k: Kernel,
) -> Result<AcoscBatchOutput, AcoscError> {
    let kernel = match k {
        Kernel::Auto => detect_best_batch_kernel(),
        other if other.is_batch() => other,
        _ => return Err(AcoscError::InvalidKernelForBatch(k)),
    };
    let simd = match kernel {
        Kernel::Avx512Batch => Kernel::Avx512,
        Kernel::Avx2Batch => Kernel::Avx2,
        Kernel::ScalarBatch => Kernel::Scalar,
        _ => unreachable!(),
    };
    acosc_batch_par_slice(high, low, simd)
}
#[inline(always)]
pub fn acosc_batch_slice(
    high: &[f64],
    low: &[f64],
    kern: Kernel,
) -> Result<AcoscBatchOutput, AcoscError> {
    acosc_batch_inner(high, low, kern, false)
}
#[inline(always)]
pub fn acosc_batch_par_slice(
    high: &[f64],
    low: &[f64],
    kern: Kernel,
) -> Result<AcoscBatchOutput, AcoscError> {
    acosc_batch_inner(high, low, kern, true)
}
#[inline(always)]
fn acosc_batch_inner(
    high: &[f64],
    low: &[f64],
    kern: Kernel,
    _parallel: bool,
) -> Result<AcoscBatchOutput, AcoscError> {
    let cols = high.len();
    let rows: usize = 1;

    let _total = rows.checked_mul(cols).ok_or(AcoscError::InvalidRange {
        start: 0,
        end: cols as i64,
        step: 0,
    })?;

    let first = (0..cols)
        .find(|&i| !high[i].is_nan() && !low[i].is_nan())
        .unwrap_or(cols);
    const REQUIRED_LENGTH: usize = 39;
    let valid = cols.saturating_sub(first);
    if valid == 0 {
        return Err(AcoscError::AllValuesNaN);
    }
    if valid < REQUIRED_LENGTH {
        return Err(AcoscError::NotEnoughValidData {
            needed: REQUIRED_LENGTH,
            valid,
        });
    }

    let mut buf_osc_mu = make_uninit_matrix(rows, cols);
    let mut buf_change_mu = make_uninit_matrix(rows, cols);

    const WARMUP: usize = 38;
    let warmups = vec![first + WARMUP];
    init_matrix_prefixes(&mut buf_osc_mu, cols, &warmups);
    init_matrix_prefixes(&mut buf_change_mu, cols, &warmups);

    let mut osc_guard = core::mem::ManuallyDrop::new(buf_osc_mu);
    let mut change_guard = core::mem::ManuallyDrop::new(buf_change_mu);

    let osc_slice: &mut [f64] = unsafe {
        core::slice::from_raw_parts_mut(osc_guard.as_mut_ptr() as *mut f64, osc_guard.len())
    };
    let change_slice: &mut [f64] = unsafe {
        core::slice::from_raw_parts_mut(change_guard.as_mut_ptr() as *mut f64, change_guard.len())
    };

    let simd = match kern {
        Kernel::Auto => detect_best_kernel(),
        other => other,
    };

    if first < cols {
        let valid_len = cols - first;
        if valid_len > WARMUP {
            acosc_compute_into(
                &high[first..],
                &low[first..],
                simd,
                &mut osc_slice[first..],
                &mut change_slice[first..],
            );
        }
    }

    let osc = unsafe {
        Vec::from_raw_parts(
            osc_guard.as_mut_ptr() as *mut f64,
            osc_guard.len(),
            osc_guard.capacity(),
        )
    };
    let change = unsafe {
        Vec::from_raw_parts(
            change_guard.as_mut_ptr() as *mut f64,
            change_guard.len(),
            change_guard.capacity(),
        )
    };

    Ok(AcoscBatchOutput {
        osc,
        change,
        rows,
        cols,
    })
}
#[inline(always)]
pub fn expand_grid(_r: &AcoscBatchRange) -> Vec<AcoscParams> {
    vec![AcoscParams::default()]
}

#[inline(always)]
pub unsafe fn acosc_row_scalar(
    high: &[f64],
    low: &[f64],
    out_osc: &mut [f64],
    out_change: &mut [f64],
) {
    acosc_scalar(high, low, out_osc, out_change)
}
#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn acosc_row_avx2(
    high: &[f64],
    low: &[f64],
    out_osc: &mut [f64],
    out_change: &mut [f64],
) {
    acosc_avx2(high, low, out_osc, out_change)
}
#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn acosc_row_avx512(
    high: &[f64],
    low: &[f64],
    out_osc: &mut [f64],
    out_change: &mut [f64],
) {
    acosc_avx512(high, low, out_osc, out_change)
}
#[inline(always)]
pub fn acosc_row_avx512_short(
    high: &[f64],
    low: &[f64],
    out_osc: &mut [f64],
    out_change: &mut [f64],
) {
    acosc_scalar(high, low, out_osc, out_change)
}
#[inline(always)]
pub fn acosc_row_avx512_long(
    high: &[f64],
    low: &[f64],
    out_osc: &mut [f64],
    out_change: &mut [f64],
) {
    acosc_scalar(high, low, out_osc, out_change)
}

#[derive(Copy, Clone, Debug, Default)]
pub struct AcoscBuilder {
    kernel: Kernel,
}
impl AcoscBuilder {
    #[inline(always)]
    pub fn new() -> Self {
        Self::default()
    }
    #[inline(always)]
    pub fn kernel(mut self, k: Kernel) -> Self {
        self.kernel = k;
        self
    }
    #[inline(always)]
    pub fn apply_candles(self, candles: &Candles) -> Result<AcoscOutput, AcoscError> {
        let input = AcoscInput::with_default_candles(candles);
        acosc_with_kernel(&input, self.kernel)
    }
    #[inline(always)]
    pub fn apply_slices(self, high: &[f64], low: &[f64]) -> Result<AcoscOutput, AcoscError> {
        let input = AcoscInput::from_slices(high, low, AcoscParams::default());
        acosc_with_kernel(&input, self.kernel)
    }
}

#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::cuda_available;
#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::oscillators::CudaAcosc;
#[cfg(all(feature = "python", feature = "cuda"))]
use crate::indicators::moving_averages::alma::DeviceArrayF32Py;
#[cfg(feature = "python")]
use numpy::{IntoPyArray, PyArray1, PyArrayMethods, PyReadonlyArray1};
#[cfg(feature = "python")]
use pyo3::exceptions::PyValueError;
#[cfg(feature = "python")]
use pyo3::prelude::*;
#[cfg(feature = "python")]
use pyo3::types::PyDict;
#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
use serde::{Deserialize, Serialize};
#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
use wasm_bindgen::prelude::*;

#[cfg(feature = "python")]
#[pyfunction(name = "acosc")]
#[pyo3(signature = (high, low, kernel=None))]

pub fn acosc_py<'py>(
    py: Python<'py>,
    high: PyReadonlyArray1<'py, f64>,
    low: PyReadonlyArray1<'py, f64>,
    kernel: Option<&str>,
) -> PyResult<(Bound<'py, PyArray1<f64>>, Bound<'py, PyArray1<f64>>)> {
    use numpy::{IntoPyArray, PyArrayMethods};

    let high_slice = high.as_slice()?;
    let low_slice = low.as_slice()?;
    let kern = crate::utilities::kernel_validation::validate_kernel(kernel, false)?;

    let params = AcoscParams::default();
    let acosc_in = AcoscInput::from_slices(high_slice, low_slice, params);

    let (osc_vec, change_vec) = py
        .allow_threads(|| {
            acosc_with_kernel(&acosc_in, kern).map(|output| (output.osc, output.change))
        })
        .map_err(|e| PyValueError::new_err(e.to_string()))?;

    Ok((osc_vec.into_pyarray(py), change_vec.into_pyarray(py)))
}

#[cfg(feature = "python")]
#[pyclass(name = "AcoscStream")]
pub struct AcoscStreamPy {
    stream: AcoscStream,
}

#[cfg(feature = "python")]
#[pymethods]
impl AcoscStreamPy {
    #[new]
    fn new() -> PyResult<Self> {
        let params = AcoscParams::default();
        let stream =
            AcoscStream::try_new(params).map_err(|e| PyValueError::new_err(e.to_string()))?;
        Ok(AcoscStreamPy { stream })
    }

    fn update(&mut self, high: f64, low: f64) -> Option<(f64, f64)> {
        self.stream.update(high, low)
    }
}

#[cfg(feature = "python")]
#[pyfunction(name = "acosc_batch")]
#[pyo3(signature = (high, low, kernel=None))]
pub fn acosc_batch_py<'py>(
    py: Python<'py>,
    high: PyReadonlyArray1<'py, f64>,
    low: PyReadonlyArray1<'py, f64>,
    kernel: Option<&str>,
) -> PyResult<Bound<'py, PyDict>> {
    use numpy::{IntoPyArray, PyArray1, PyArrayMethods};

    let h = high.as_slice()?;
    let l = low.as_slice()?;
    let kern = crate::utilities::kernel_validation::validate_kernel(kernel, true)?;

    let rows = 1usize;
    let cols = h.len();

    let out_osc = unsafe { PyArray1::<f64>::new(py, [rows * cols], false) };
    let out_change = unsafe { PyArray1::<f64>::new(py, [rows * cols], false) };
    let slice_osc = unsafe { out_osc.as_slice_mut()? };
    let slice_change = unsafe { out_change.as_slice_mut()? };

    py.allow_threads(|| -> Result<(), AcoscError> {
        let simd = match kern {
            Kernel::Auto => detect_best_batch_kernel(),
            k => k,
        };
        let simd = match simd {
            Kernel::Avx512Batch => Kernel::Avx512,
            Kernel::Avx2Batch => Kernel::Avx2,
            Kernel::ScalarBatch => Kernel::Scalar,
            _ => simd,
        };

        let first = (0..cols)
            .find(|&i| !h[i].is_nan() && !l[i].is_nan())
            .unwrap_or(cols);
        const REQUIRED_LENGTH: usize = 39;
        let valid = cols.saturating_sub(first);
        if valid < REQUIRED_LENGTH {
            return Err(AcoscError::NotEnoughValidData {
                needed: REQUIRED_LENGTH,
                valid,
            });
        }

        const WARMUP: usize = 38;
        let warm = first + WARMUP;

        for i in 0..warm.min(cols) {
            slice_osc[i] = f64::from_bits(0x7ff8_0000_0000_0000);
            slice_change[i] = f64::from_bits(0x7ff8_0000_0000_0000);
        }

        if first < cols && valid > WARMUP {
            acosc_compute_into(
                &h[first..],
                &l[first..],
                simd,
                &mut slice_osc[first..],
                &mut slice_change[first..],
            )
        };
        Ok(())
    })
    .map_err(|e| PyValueError::new_err(e.to_string()))?;

    let d = PyDict::new(py);
    d.set_item("osc", out_osc.reshape((rows, cols))?)?;
    d.set_item("change", out_change.reshape((rows, cols))?)?;
    Ok(d)
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyfunction(name = "acosc_cuda_batch_dev")]
#[pyo3(signature = (high_f32, low_f32, device_id=0))]
pub fn acosc_cuda_batch_dev_py(
    py: Python<'_>,
    high_f32: numpy::PyReadonlyArray1<'_, f32>,
    low_f32: numpy::PyReadonlyArray1<'_, f32>,
    device_id: usize,
) -> PyResult<(AcoscDeviceArrayF32Py, AcoscDeviceArrayF32Py)> {
    if !cuda_available() {
        return Err(PyValueError::new_err("CUDA not available"));
    }
    let h = high_f32.as_slice()?;
    let l = low_f32.as_slice()?;
    let pair = py.allow_threads(|| {
        let cuda = CudaAcosc::new(device_id).map_err(|e| PyValueError::new_err(e.to_string()))?;
        cuda.acosc_batch_dev(h, l)
            .map_err(|e| PyValueError::new_err(e.to_string()))
    })?;
    Ok((
        AcoscDeviceArrayF32Py {
            inner: Some(pair.osc),
            device_id: device_id as u32,
        },
        AcoscDeviceArrayF32Py {
            inner: Some(pair.change),
            device_id: device_id as u32,
        },
    ))
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyfunction(name = "acosc_cuda_many_series_one_param_dev")]
#[pyo3(signature = (high_tm_f32, low_tm_f32, device_id=0))]
pub fn acosc_cuda_many_series_one_param_dev_py(
    py: Python<'_>,
    high_tm_f32: numpy::PyReadonlyArray2<'_, f32>,
    low_tm_f32: numpy::PyReadonlyArray2<'_, f32>,
    device_id: usize,
) -> PyResult<(AcoscDeviceArrayF32Py, AcoscDeviceArrayF32Py)> {
    use numpy::PyUntypedArrayMethods;
    if !cuda_available() {
        return Err(PyValueError::new_err("CUDA not available"));
    }
    let shape_h = high_tm_f32.shape();
    let shape_l = low_tm_f32.shape();
    if shape_h != shape_l || shape_h.len() != 2 {
        return Err(PyValueError::new_err("high/low must be same 2D shape"));
    }
    let rows = shape_h[0];
    let cols = shape_h[1];
    let h = high_tm_f32.as_slice()?;
    let l = low_tm_f32.as_slice()?;
    let pair = py.allow_threads(|| {
        let cuda = CudaAcosc::new(device_id).map_err(|e| PyValueError::new_err(e.to_string()))?;
        cuda.acosc_many_series_one_param_time_major_dev(h, l, cols, rows)
            .map_err(|e| PyValueError::new_err(e.to_string()))
    })?;
    Ok((
        AcoscDeviceArrayF32Py {
            inner: Some(pair.osc),
            device_id: device_id as u32,
        },
        AcoscDeviceArrayF32Py {
            inner: Some(pair.change),
            device_id: device_id as u32,
        },
    ))
}

#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::oscillators::DeviceArrayF32Acosc;
#[cfg(all(feature = "python", feature = "cuda"))]
#[pyclass(module = "ta_indicators.cuda", unsendable)]
pub struct AcoscDeviceArrayF32Py {
    pub(crate) inner: Option<DeviceArrayF32Acosc>,
    pub(crate) device_id: u32,
}
#[cfg(all(feature = "python", feature = "cuda"))]
#[pymethods]
impl AcoscDeviceArrayF32Py {
    #[getter]
    fn __cuda_array_interface__<'py>(&self, py: Python<'py>) -> PyResult<Bound<'py, PyDict>> {
        let inner = self
            .inner
            .as_ref()
            .ok_or_else(|| PyValueError::new_err("buffer already exported via __dlpack__"))?;
        let d = PyDict::new(py);
        d.set_item("shape", (inner.rows, inner.cols))?;
        d.set_item("typestr", "<f4")?;
        d.set_item(
            "strides",
            (
                inner.cols * std::mem::size_of::<f32>(),
                std::mem::size_of::<f32>(),
            ),
        )?;
        d.set_item("data", (inner.device_ptr() as usize, false))?;

        d.set_item("version", 3)?;
        Ok(d)
    }

    fn __dlpack_device__(&self) -> PyResult<(i32, i32)> {
        let inner = self
            .inner
            .as_ref()
            .ok_or_else(|| PyValueError::new_err("buffer already exported via __dlpack__"))?;
        Ok((2, inner.device_id as i32))
    }

    fn __dlpack__<'py>(
        &mut self,
        py: Python<'py>,
        stream: Option<i64>,
        max_version: Option<(u32, u32)>,
        dl_device: Option<(i32, i32)>,
        _copy: Option<bool>,
    ) -> PyResult<PyObject> {
        use crate::utilities::dlpack_cuda::export_f32_cuda_dlpack_2d;

        let inner = self
            .inner
            .take()
            .ok_or_else(|| PyValueError::new_err("buffer already exported via __dlpack__"))?;

        if let Some((_ty, dev_id)) = dl_device {
            if dev_id as u32 != inner.device_id {
                return Err(PyValueError::new_err(
                    "dl_device does not match allocation device",
                ));
            }
        }

        let _ = stream;

        let DeviceArrayF32Acosc {
            buf,
            rows,
            cols,
            ctx: _,
            device_id,
        } = inner;

        let max_version_bound = max_version
            .map(|(maj, min)| -> PyResult<_> {
                use pyo3::IntoPyObjectExt;
                (maj as i32, min as i32).into_bound_py_any(py)
            })
            .transpose()?;

        export_f32_cuda_dlpack_2d(py, buf, rows, cols, device_id as i32, max_version_bound)
    }
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn acosc_js(high: &[f64], low: &[f64]) -> Result<Vec<f64>, JsValue> {
    let params = AcoscParams::default();
    let input = AcoscInput::from_slices(high, low, params);

    let len = high.len();
    let total = len
        .checked_mul(2)
        .ok_or_else(|| JsValue::from_str("acosc_js: size overflow"))?;
    let mut output = vec![0.0; total];

    let (osc_slice, change_slice) = output.split_at_mut(len);

    acosc_into_slice(osc_slice, change_slice, &input, Kernel::Auto)
        .map_err(|e| JsValue::from_str(&e.to_string()))?;

    Ok(output)
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[derive(Serialize, Deserialize)]
pub struct AcoscBatchConfig {}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[derive(Serialize, Deserialize)]
pub struct AcoscBatchJsOutput {
    pub values: Vec<f64>,
    pub rows: usize,
    pub cols: usize,
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen(js_name = acosc_batch)]
pub fn acosc_batch_unified_js(
    high: &[f64],
    low: &[f64],
    _config: JsValue,
) -> Result<JsValue, JsValue> {
    let rows = 1;
    let cols = high.len();

    let total = cols
        .checked_mul(2)
        .ok_or_else(|| JsValue::from_str("acosc_batch_unified_js: size overflow"))?;
    let mut output = vec![0.0; total];

    let (osc_slice, change_slice) = output.split_at_mut(cols);

    let params = AcoscParams::default();
    let input = AcoscInput::from_slices(high, low, params);

    acosc_into_slice(osc_slice, change_slice, &input, Kernel::Auto)
        .map_err(|e| JsValue::from_str(&e.to_string()))?;

    let js_output = AcoscBatchJsOutput {
        values: output,
        rows,
        cols,
    };

    serde_wasm_bindgen::to_value(&js_output)
        .map_err(|e| JsValue::from_str(&format!("Serialization error: {}", e)))
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn acosc_batch_js(high: &[f64], low: &[f64]) -> Result<Vec<f64>, JsValue> {
    acosc_js(high, low)
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn acosc_batch_metadata_js() -> Result<Vec<f64>, JsValue> {
    Ok(vec![])
}

pub fn acosc_into_slice(
    osc_dst: &mut [f64],
    change_dst: &mut [f64],
    input: &AcoscInput,
    kern: Kernel,
) -> Result<(), AcoscError> {
    let (high, low, first, kernel) = acosc_prepare(input, kern)?;

    if osc_dst.len() != high.len() {
        return Err(AcoscError::OutputLengthMismatch {
            expected: high.len(),
            got: osc_dst.len(),
        });
    }
    if change_dst.len() != high.len() {
        return Err(AcoscError::OutputLengthMismatch {
            expected: high.len(),
            got: change_dst.len(),
        });
    }

    const WARMUP: usize = 38;
    let warm = first + WARMUP;
    for i in 0..warm.min(osc_dst.len()) {
        osc_dst[i] = f64::from_bits(0x7ff8_0000_0000_0000);
        change_dst[i] = f64::from_bits(0x7ff8_0000_0000_0000);
    }

    let valid = high.len() - first;
    if first < high.len() && valid > WARMUP {
        acosc_compute_into(
            &high[first..],
            &low[first..],
            kernel,
            &mut osc_dst[first..],
            &mut change_dst[first..],
        );
    }
    Ok(())
}

#[cfg(not(all(target_arch = "wasm32", feature = "wasm")))]
#[inline]
pub fn acosc_into(
    input: &AcoscInput,
    osc_out: &mut [f64],
    change_out: &mut [f64],
) -> Result<(), AcoscError> {
    acosc_into_slice(osc_out, change_out, input, Kernel::Auto)
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn acosc_into(
    high_ptr: *const f64,
    low_ptr: *const f64,
    osc_ptr: *mut f64,
    change_ptr: *mut f64,
    len: usize,
) -> Result<(), JsValue> {
    if high_ptr.is_null() || low_ptr.is_null() || osc_ptr.is_null() || change_ptr.is_null() {
        return Err(JsValue::from_str("null pointer passed to acosc_into"));
    }

    unsafe {
        let high = std::slice::from_raw_parts(high_ptr, len);
        let low = std::slice::from_raw_parts(low_ptr, len);

        if len < 39 {
            return Err(JsValue::from_str("Not enough data"));
        }

        let params = AcoscParams::default();
        let input = AcoscInput::from_slices(high, low, params);

        let need_temp = high_ptr == osc_ptr as *const f64
            || high_ptr == change_ptr as *const f64
            || low_ptr == osc_ptr as *const f64
            || low_ptr == change_ptr as *const f64
            || osc_ptr == change_ptr;

        if need_temp {
            let mut temp_osc = vec![0.0; len];
            let mut temp_change = vec![0.0; len];

            acosc_into_slice(&mut temp_osc, &mut temp_change, &input, Kernel::Auto)
                .map_err(|e| JsValue::from_str(&e.to_string()))?;

            let osc_out = std::slice::from_raw_parts_mut(osc_ptr, len);
            let change_out = std::slice::from_raw_parts_mut(change_ptr, len);
            osc_out.copy_from_slice(&temp_osc);
            change_out.copy_from_slice(&temp_change);
        } else {
            let osc_out = std::slice::from_raw_parts_mut(osc_ptr, len);
            let change_out = std::slice::from_raw_parts_mut(change_ptr, len);

            acosc_into_slice(osc_out, change_out, &input, Kernel::Auto)
                .map_err(|e| JsValue::from_str(&e.to_string()))?;
        }

        Ok(())
    }
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn acosc_alloc(len: usize) -> *mut f64 {
    let mut vec = Vec::<f64>::with_capacity(len);
    let ptr = vec.as_mut_ptr();
    std::mem::forget(vec);
    ptr
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn acosc_free(ptr: *mut f64, len: usize) {
    if !ptr.is_null() {
        unsafe {
            let _ = Vec::from_raw_parts(ptr, len, len);
        }
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::skip_if_unsupported;
    use crate::utilities::data_loader::read_candles_from_csv;
    use std::error::Error;

    fn check_acosc_partial_params(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;
        let default_params = AcoscParams::default();
        let input = AcoscInput::from_candles(&candles, default_params);
        let output = acosc_with_kernel(&input, kernel)?;
        assert_eq!(output.osc.len(), candles.close.len());
        assert_eq!(output.change.len(), candles.close.len());
        Ok(())
    }

    fn check_acosc_accuracy(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;
        let input = AcoscInput::with_default_candles(&candles);
        let result = acosc_with_kernel(&input, kernel)?;
        assert_eq!(result.osc.len(), candles.close.len());
        assert_eq!(result.change.len(), candles.close.len());
        let expected_last_five_acosc_osc = [273.30, 383.72, 357.7, 291.25, 176.84];
        let expected_last_five_acosc_change = [49.6, 110.4, -26.0, -66.5, -114.4];
        let start = result.osc.len().saturating_sub(5);
        for (i, &val) in result.osc[start..].iter().enumerate() {
            assert!(
                (val - expected_last_five_acosc_osc[i]).abs() < 1e-1,
                "[{}] ACOSC {:?} osc mismatch idx {}: got {}, expected {}",
                test_name,
                kernel,
                i,
                val,
                expected_last_five_acosc_osc[i]
            );
        }
        for (i, &val) in result.change[start..].iter().enumerate() {
            assert!(
                (val - expected_last_five_acosc_change[i]).abs() < 1e-1,
                "[{}] ACOSC {:?} change mismatch idx {}: got {}, expected {}",
                test_name,
                kernel,
                i,
                val,
                expected_last_five_acosc_change[i]
            );
        }
        Ok(())
    }

    fn check_acosc_default_candles(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;
        let input = AcoscInput::with_default_candles(&candles);
        match input.data {
            AcoscData::Candles { .. } => {}
            _ => panic!("Expected AcoscData::Candles variant"),
        }
        let output = acosc_with_kernel(&input, kernel)?;
        assert_eq!(output.osc.len(), candles.close.len());
        Ok(())
    }

    fn check_acosc_too_short(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let high = [100.0, 101.0];
        let low = [99.0, 98.0];
        let params = AcoscParams::default();
        let input = AcoscInput::from_slices(&high, &low, params);
        let result = acosc_with_kernel(&input, kernel);
        assert!(
            result.is_err(),
            "[{}] Should fail with not enough data",
            test_name
        );
        Ok(())
    }

    fn check_acosc_reinput(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;
        let input = AcoscInput::with_default_candles(&candles);
        let first_result = acosc_with_kernel(&input, kernel)?;
        assert_eq!(first_result.osc.len(), candles.close.len());
        assert_eq!(first_result.change.len(), candles.close.len());
        let input2 = AcoscInput::from_slices(&candles.high, &candles.low, AcoscParams::default());
        let second_result = acosc_with_kernel(&input2, kernel)?;
        assert_eq!(second_result.osc.len(), candles.close.len());
        for (a, b) in second_result.osc.iter().zip(first_result.osc.iter()) {
            if a.is_nan() && b.is_nan() {
                continue;
            }
            assert!(
                (a - b).abs() < 1e-8,
                "Reinput values mismatch: {} vs {}",
                a,
                b
            );
        }
        Ok(())
    }

    fn check_acosc_nan_handling(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;
        let input = AcoscInput::with_default_candles(&candles);
        let result = acosc_with_kernel(&input, kernel)?;
        if result.osc.len() > 240 {
            for i in 240..result.osc.len() {
                assert!(!result.osc[i].is_nan(), "Found NaN in osc at {}", i);
                assert!(!result.change[i].is_nan(), "Found NaN in change at {}", i);
            }
        }
        Ok(())
    }

    fn check_acosc_streaming(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;
        let input = AcoscInput::with_default_candles(&candles);
        let batch = acosc_with_kernel(&input, kernel)?;
        let mut stream = AcoscStream::try_new(AcoscParams::default())?;
        let mut osc_stream = Vec::with_capacity(candles.close.len());
        let mut change_stream = Vec::with_capacity(candles.close.len());
        for (&h, &l) in candles.high.iter().zip(candles.low.iter()) {
            match stream.update(h, l) {
                Some((o, c)) => {
                    osc_stream.push(o);
                    change_stream.push(c);
                }
                None => {
                    osc_stream.push(f64::NAN);
                    change_stream.push(f64::NAN);
                }
            }
        }
        assert_eq!(batch.osc.len(), osc_stream.len());
        assert_eq!(batch.change.len(), change_stream.len());
        for (i, (&a, &b)) in batch.osc.iter().zip(osc_stream.iter()).enumerate() {
            if a.is_nan() && b.is_nan() {
                continue;
            }
            assert!(
                (a - b).abs() < 1e-9,
                "Streaming osc mismatch at idx {}: {} vs {}",
                i,
                a,
                b
            );
        }
        for (i, (&a, &b)) in batch.change.iter().zip(change_stream.iter()).enumerate() {
            if a.is_nan() && b.is_nan() {
                continue;
            }
            assert!(
                (a - b).abs() < 1e-9,
                "Streaming change mismatch at idx {}: {} vs {}",
                i,
                a,
                b
            );
        }
        Ok(())
    }

    macro_rules! generate_all_acosc_tests {
        ($($test_fn:ident),*) => {
            paste::paste! {
                $(#[test]
                  fn [<$test_fn _scalar_f64>]() {
                      let _ = $test_fn(stringify!([<$test_fn _scalar_f64>]), Kernel::Scalar);
                  })*
                #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
                $(#[test]
                  fn [<$test_fn _avx2_f64>]() {
                      let _ = $test_fn(stringify!([<$test_fn _avx2_f64>]), Kernel::Avx2);
                  }
                  #[test]
                  fn [<$test_fn _avx512_f64>]() {
                      let _ = $test_fn(stringify!([<$test_fn _avx512_f64>]), Kernel::Avx512);
                  })*
            }
        }
    }
    generate_all_acosc_tests!(
        check_acosc_partial_params,
        check_acosc_accuracy,
        check_acosc_default_candles,
        check_acosc_too_short,
        check_acosc_reinput,
        check_acosc_nan_handling,
        check_acosc_streaming,
        check_acosc_no_poison
    );

    #[cfg(feature = "proptest")]
    generate_all_acosc_tests!(check_acosc_property);

    #[cfg(feature = "proptest")]
    fn check_acosc_property(
        test_name: &str,
        kernel: Kernel,
    ) -> Result<(), Box<dyn std::error::Error>> {
        use proptest::prelude::*;
        skip_if_unsupported!(kernel, test_name);

        let strat = (40usize..=400).prop_flat_map(|len| {
            prop::collection::vec(
                (1.0f64..10000.0f64)
                    .prop_flat_map(|base_price| {
                        (0.0f64..0.1f64).prop_map(move |spread_pct| {
                            let half_spread = base_price * spread_pct * 0.5;
                            let high = base_price + half_spread;
                            let low = base_price - half_spread;
                            (high, low)
                        })
                    })
                    .prop_filter("prices must be finite", |(h, l)| {
                        h.is_finite() && l.is_finite()
                    }),
                len,
            )
        });

        proptest::test_runner::TestRunner::default().run(&strat, |price_pairs| {
            let (high_vec, low_vec): (Vec<f64>, Vec<f64>) = price_pairs.into_iter().unzip();
            let params = AcoscParams::default();
            let input = AcoscInput::from_slices(&high_vec, &low_vec, params);

            let result = acosc_with_kernel(&input, kernel).unwrap();
            let scalar_result = acosc_with_kernel(&input, Kernel::Scalar).unwrap();

            for i in 0..result.osc.len() {
                let y = result.osc[i];
                let r = scalar_result.osc[i];

                if !y.is_finite() || !r.is_finite() {
                    prop_assert_eq!(
                        y.to_bits(),
                        r.to_bits(),
                        "NaN/finite mismatch in osc at idx {}: {} vs {}",
                        i,
                        y,
                        r
                    );
                    continue;
                }

                let y_bits = y.to_bits();
                let r_bits = r.to_bits();
                let ulp_diff: u64 = y_bits.abs_diff(r_bits);

                prop_assert!(
                    (y - r).abs() <= 1e-9 || ulp_diff <= 4,
                    "Kernel mismatch in osc at idx {}: {} vs {} (ULP={})",
                    i,
                    y,
                    r,
                    ulp_diff
                );
            }

            for i in 0..result.change.len() {
                let y = result.change[i];
                let r = scalar_result.change[i];

                if !y.is_finite() || !r.is_finite() {
                    prop_assert_eq!(
                        y.to_bits(),
                        r.to_bits(),
                        "NaN/finite mismatch in change at idx {}: {} vs {}",
                        i,
                        y,
                        r
                    );
                    continue;
                }

                let y_bits = y.to_bits();
                let r_bits = r.to_bits();
                let ulp_diff: u64 = y_bits.abs_diff(r_bits);

                prop_assert!(
                    (y - r).abs() <= 1e-9 || ulp_diff <= 4,
                    "Kernel mismatch in change at idx {}: {} vs {} (ULP={})",
                    i,
                    y,
                    r,
                    ulp_diff
                );
            }

            for i in 0..38.min(result.osc.len()) {
                prop_assert!(
                    result.osc[i].is_nan(),
                    "Expected NaN in osc warmup at idx {}, got {}",
                    i,
                    result.osc[i]
                );
                prop_assert!(
                    result.change[i].is_nan(),
                    "Expected NaN in change warmup at idx {}, got {}",
                    i,
                    result.change[i]
                );
            }

            if result.osc.len() > 38 {
                prop_assert!(
                    result.osc[38].is_finite(),
                    "Expected finite value at idx 38 in osc, got {}",
                    result.osc[38]
                );
                prop_assert!(
                    result.change[38].is_finite(),
                    "Expected finite value at idx 38 in change, got {}",
                    result.change[38]
                );
            }

            for i in 39..result.osc.len() {
                if result.osc[i].is_finite() && result.osc[i - 1].is_finite() {
                    let expected_change = result.osc[i] - result.osc[i - 1];
                    let actual_change = result.change[i];

                    prop_assert!(
                        (expected_change - actual_change).abs() <= 1e-9,
                        "Change formula mismatch at idx {}: expected {} ({}−{}), got {}",
                        i,
                        expected_change,
                        result.osc[i],
                        result.osc[i - 1],
                        actual_change
                    );
                }
            }

            if high_vec.windows(2).all(|w| (w[0] - w[1]).abs() < 1e-10)
                && low_vec.windows(2).all(|w| (w[0] - w[1]).abs() < 1e-10)
            {
                for i in 39..result.osc.len() {
                    prop_assert!(
                        result.osc[i].abs() <= 1e-6,
                        "Expected near-zero osc with constant prices at idx {}, got {}",
                        i,
                        result.osc[i]
                    );
                }
            }

            Ok(())
        })?;

        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;
        let test_len = candles.high.len().min(200);
        let high_data = &candles.high[..test_len];
        let low_data = &candles.low[..test_len];

        {
            let params = AcoscParams::default();
            let input = AcoscInput::from_slices(high_data, low_data, params.clone());
            let batch_result = acosc_with_kernel(&input, kernel)?;

            let mut stream = AcoscStream::try_new(params)?;
            let mut stream_osc = Vec::with_capacity(test_len);
            let mut stream_change = Vec::with_capacity(test_len);

            for i in 0..test_len {
                match stream.update(high_data[i], low_data[i]) {
                    Some((osc, change)) => {
                        stream_osc.push(osc);
                        stream_change.push(change);
                    }
                    None => {
                        stream_osc.push(f64::NAN);
                        stream_change.push(f64::NAN);
                    }
                }
            }

            for i in 0..test_len {
                let batch_o = batch_result.osc[i];
                let stream_o = stream_osc[i];

                if batch_o.is_nan() && stream_o.is_nan() {
                    continue;
                }

                assert!(
                    (batch_o - stream_o).abs() <= 1e-9,
                    "[{}] Streaming vs batch mismatch in osc at idx {}: {} vs {}",
                    test_name,
                    i,
                    batch_o,
                    stream_o
                );

                let batch_c = batch_result.change[i];
                let stream_c = stream_change[i];

                if batch_c.is_nan() && stream_c.is_nan() {
                    continue;
                }

                assert!(
                    (batch_c - stream_c).abs() <= 1e-9,
                    "[{}] Streaming vs batch mismatch in change at idx {}: {} vs {}",
                    test_name,
                    i,
                    batch_c,
                    stream_c
                );
            }
        }

        Ok(())
    }

    fn check_batch_default_row(test: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test);
        let file = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let c = read_candles_from_csv(file)?;
        let output = AcoscBatchBuilder::new().kernel(kernel).apply_candles(&c)?;
        assert_eq!(output.osc.len(), c.close.len());
        Ok(())
    }

    #[cfg(debug_assertions)]
    fn check_acosc_no_poison(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;
        let input = AcoscInput::with_default_candles(&candles);
        let output = acosc_with_kernel(&input, kernel)?;

        for (i, &val) in output.osc.iter().enumerate() {
            if val.is_nan() {
                continue;
            }

            let bits = val.to_bits();

            if bits == 0x11111111_11111111 {
                panic!(
					"[{}] Found alloc_with_nan_prefix poison value {} (0x{:016X}) at index {} in osc",
					test_name, val, bits, i
				);
            }
        }

        for (i, &val) in output.change.iter().enumerate() {
            if val.is_nan() {
                continue;
            }

            let bits = val.to_bits();

            if bits == 0x11111111_11111111 {
                panic!(
					"[{}] Found alloc_with_nan_prefix poison value {} (0x{:016X}) at index {} in change",
					test_name, val, bits, i
				);
            }
        }

        Ok(())
    }

    #[cfg(debug_assertions)]
    fn check_batch_no_poison(test: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test);
        let file = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let c = read_candles_from_csv(file)?;
        let output = AcoscBatchBuilder::new().kernel(kernel).apply_candles(&c)?;

        for (idx, &val) in output.osc.iter().enumerate() {
            if val.is_nan() {
                continue;
            }

            let bits = val.to_bits();

            if bits == 0x11111111_11111111 {
                panic!(
					"[{}] Found alloc_with_nan_prefix poison value {} (0x{:016X}) at index {} in osc",
					test, val, bits, idx
				);
            }

            if bits == 0x22222222_22222222 {
                panic!(
					"[{}] Found init_matrix_prefixes poison value {} (0x{:016X}) at index {} in osc",
					test, val, bits, idx
				);
            }

            if bits == 0x33333333_33333333 {
                panic!(
                    "[{}] Found make_uninit_matrix poison value {} (0x{:016X}) at index {} in osc",
                    test, val, bits, idx
                );
            }
        }

        for (idx, &val) in output.change.iter().enumerate() {
            if val.is_nan() {
                continue;
            }

            let bits = val.to_bits();

            if bits == 0x11111111_11111111 {
                panic!(
					"[{}] Found alloc_with_nan_prefix poison value {} (0x{:016X}) at index {} in change",
					test, val, bits, idx
				);
            }

            if bits == 0x22222222_22222222 {
                panic!(
					"[{}] Found init_matrix_prefixes poison value {} (0x{:016X}) at index {} in change",
					test, val, bits, idx
				);
            }

            if bits == 0x33333333_33333333 {
                panic!(
					"[{}] Found make_uninit_matrix poison value {} (0x{:016X}) at index {} in change",
					test, val, bits, idx
				);
            }
        }

        Ok(())
    }

    #[cfg(not(debug_assertions))]
    fn check_acosc_no_poison(_test_name: &str, _kernel: Kernel) -> Result<(), Box<dyn Error>> {
        Ok(())
    }

    #[cfg(not(debug_assertions))]
    fn check_batch_no_poison(_test: &str, _kernel: Kernel) -> Result<(), Box<dyn Error>> {
        Ok(())
    }
    macro_rules! gen_batch_tests {
        ($fn_name:ident) => {
            paste::paste! {
                #[test] fn [<$fn_name _scalar>]()      {
                    let _ = $fn_name(stringify!([<$fn_name _scalar>]), Kernel::ScalarBatch);
                }
                #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
                #[test] fn [<$fn_name _avx2>]()        {
                    let _ = $fn_name(stringify!([<$fn_name _avx2>]), Kernel::Avx2Batch);
                }
                #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
                #[test] fn [<$fn_name _avx512>]()      {
                    let _ = $fn_name(stringify!([<$fn_name _avx512>]), Kernel::Avx512Batch);
                }
                #[test] fn [<$fn_name _auto_detect>]() {
                    let _ = $fn_name(stringify!([<$fn_name _auto_detect>]), Kernel::Auto);
                }
            }
        };
    }
    gen_batch_tests!(check_batch_default_row);
    gen_batch_tests!(check_batch_no_poison);

    #[test]
    fn test_batch_kernel_error() {
        let high = vec![100.0; 50];
        let low = vec![99.0; 50];

        let result = acosc_batch_with_kernel(&high, &low, Kernel::Scalar);
        assert!(result.is_err());

        match result.unwrap_err() {
            AcoscError::InvalidKernelForBatch(kernel) => {
                assert_eq!(kernel, Kernel::Scalar);
            }
            _ => panic!("Expected InvalidKernelForBatch error"),
        }

        let result = acosc_batch_with_kernel(&high, &low, Kernel::Avx2);
        assert!(matches!(
            result,
            Err(AcoscError::InvalidKernelForBatch(Kernel::Avx2))
        ));
    }

    #[test]
    fn test_acosc_into_matches_api() -> Result<(), Box<dyn Error>> {
        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;
        let n = candles.high.len().min(512).max(64);
        let high = &candles.high[..n];
        let low = &candles.low[..n];

        let params = AcoscParams::default();
        let input = AcoscInput::from_slices(high, low, params);

        let base = acosc(&input)?;

        let mut out_osc = vec![0.0; n];
        let mut out_change = vec![0.0; n];

        acosc_into(&input, &mut out_osc, &mut out_change)?;

        assert_eq!(base.osc.len(), out_osc.len());
        assert_eq!(base.change.len(), out_change.len());

        fn eq_or_both_nan(a: f64, b: f64) -> bool {
            (a.is_nan() && b.is_nan()) || (a == b)
        }

        for i in 0..n {
            assert!(
                eq_or_both_nan(base.osc[i], out_osc[i]),
                "osc mismatch at {}: base={} out={}",
                i,
                base.osc[i],
                out_osc[i]
            );
            assert!(
                eq_or_both_nan(base.change[i], out_change[i]),
                "change mismatch at {}: base={} out={}",
                i,
                base.change[i],
                out_change[i]
            );
        }

        Ok(())
    }
}