vector_ta/indicators/moving_averages/

tilson.rs

#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::cuda_available;
#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::moving_averages::tilson_wrapper::DeviceArrayF32Tilson;
#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::moving_averages::{CudaTilson, CudaTilsonError};
use crate::utilities::data_loader::{source_type, 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(feature = "python")]
use crate::utilities::kernel_validation::validate_kernel;
#[cfg(all(feature = "python", feature = "cuda"))]
use cust::memory::DeviceBuffer;

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
use core::arch::x86_64::*;
#[cfg(not(target_arch = "wasm32"))]
use rayon::prelude::*;
use std::convert::AsRef;
use std::error::Error;
use std::mem::MaybeUninit;
use thiserror::Error;

#[cfg(feature = "python")]
use numpy::{IntoPyArray, PyArray1};
#[cfg(feature = "python")]
use pyo3::exceptions::PyValueError;
#[cfg(feature = "python")]
use pyo3::prelude::*;
#[cfg(feature = "python")]
use pyo3::types::{PyDict, PyList};
#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
use serde::{Deserialize, Serialize};
#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
use wasm_bindgen::prelude::*;

impl<'a> AsRef<[f64]> for TilsonInput<'a> {
    #[inline(always)]
    fn as_ref(&self) -> &[f64] {
        match &self.data {
            TilsonData::Slice(slice) => slice,
            TilsonData::Candles { candles, source } => source_type(candles, source),
        }
    }
}

#[derive(Debug, Clone)]
pub enum TilsonData<'a> {
    Candles {
        candles: &'a Candles,
        source: &'a str,
    },
    Slice(&'a [f64]),
}

#[derive(Debug, Clone)]
pub struct TilsonOutput {
    pub values: Vec<f64>,
}

#[derive(Debug, Clone)]
#[cfg_attr(
    all(target_arch = "wasm32", feature = "wasm"),
    derive(Serialize, Deserialize)
)]
pub struct TilsonParams {
    pub period: Option<usize>,
    pub volume_factor: Option<f64>,
}

impl Default for TilsonParams {
    fn default() -> Self {
        Self {
            period: Some(5),
            volume_factor: Some(0.0),
        }
    }
}

#[derive(Debug, Clone)]
pub struct TilsonInput<'a> {
    pub data: TilsonData<'a>,
    pub params: TilsonParams,
}

impl<'a> TilsonInput<'a> {
    #[inline]
    pub fn from_candles(c: &'a Candles, s: &'a str, p: TilsonParams) -> Self {
        Self {
            data: TilsonData::Candles {
                candles: c,
                source: s,
            },
            params: p,
        }
    }
    #[inline]
    pub fn from_slice(sl: &'a [f64], p: TilsonParams) -> Self {
        Self {
            data: TilsonData::Slice(sl),
            params: p,
        }
    }
    #[inline]
    pub fn with_default_candles(c: &'a Candles) -> Self {
        Self::from_candles(c, "close", TilsonParams::default())
    }
    #[inline]
    pub fn get_period(&self) -> usize {
        self.params.period.unwrap_or(5)
    }
    #[inline]
    pub fn get_volume_factor(&self) -> f64 {
        self.params.volume_factor.unwrap_or(0.0)
    }
}

#[derive(Copy, Clone, Debug)]
pub struct TilsonBuilder {
    period: Option<usize>,
    volume_factor: Option<f64>,
    kernel: Kernel,
}

impl Default for TilsonBuilder {
    fn default() -> Self {
        Self {
            period: None,
            volume_factor: None,
            kernel: Kernel::Auto,
        }
    }
}

impl TilsonBuilder {
    #[inline(always)]
    pub fn new() -> Self {
        Self::default()
    }
    #[inline(always)]
    pub fn period(mut self, n: usize) -> Self {
        self.period = Some(n);
        self
    }
    #[inline(always)]
    pub fn volume_factor(mut self, v: f64) -> Self {
        self.volume_factor = Some(v);
        self
    }
    #[inline(always)]
    pub fn kernel(mut self, k: Kernel) -> Self {
        self.kernel = k;
        self
    }

    #[inline(always)]
    pub fn apply(self, c: &Candles) -> Result<TilsonOutput, TilsonError> {
        let p = TilsonParams {
            period: self.period,
            volume_factor: self.volume_factor,
        };
        let i = TilsonInput::from_candles(c, "close", p);
        tilson_with_kernel(&i, self.kernel)
    }

    #[inline(always)]
    pub fn apply_slice(self, d: &[f64]) -> Result<TilsonOutput, TilsonError> {
        let p = TilsonParams {
            period: self.period,
            volume_factor: self.volume_factor,
        };
        let i = TilsonInput::from_slice(d, p);
        tilson_with_kernel(&i, self.kernel)
    }

    #[inline(always)]
    pub fn into_stream(self) -> Result<TilsonStream, TilsonError> {
        let p = TilsonParams {
            period: self.period,
            volume_factor: self.volume_factor,
        };
        TilsonStream::try_new(p)
    }
}

#[derive(Debug, Error)]
pub enum TilsonError {
    #[error("tilson: Input data slice is empty.")]
    EmptyInputData,

    #[error("tilson: All values are NaN.")]
    AllValuesNaN,

    #[error("tilson: Invalid period: period = {period}, data length = {data_len}")]
    InvalidPeriod { period: usize, data_len: usize },

    #[error("tilson: Not enough valid data: needed = {needed}, valid = {valid}")]
    NotEnoughValidData { needed: usize, valid: usize },

    #[error("tilson: Invalid volume factor: {v_factor}")]
    InvalidVolumeFactor { v_factor: f64 },

    #[error("tilson: Output length mismatch: expected = {expected}, got = {got}")]
    OutputLengthMismatch { expected: usize, got: usize },

    #[error("tilson: Invalid kernel for batch operation: {0:?}")]
    InvalidKernelForBatch(Kernel),

    #[error("tilson: Invalid integer range expansion: start={start}, end={end}, step={step}")]
    InvalidRangeUsize {
        start: usize,
        end: usize,
        step: usize,
    },

    #[error("tilson: Invalid float range expansion: start={start}, end={end}, step={step}")]
    InvalidRangeF64 { start: f64, end: f64, step: f64 },

    #[error("tilson: invalid input: {0}")]
    InvalidInput(&'static str),
}

#[inline]
pub fn tilson(input: &TilsonInput) -> Result<TilsonOutput, TilsonError> {
    tilson_with_kernel(input, Kernel::Auto)
}

pub fn tilson_with_kernel(
    input: &TilsonInput,
    kernel: Kernel,
) -> Result<TilsonOutput, TilsonError> {
    let (data, period, v_factor, first, len, chosen) = tilson_prepare(input, kernel)?;
    let lookback_total = 6 * (period - 1);
    let warm = first + lookback_total;

    let mut out = alloc_with_nan_prefix(len, warm);
    tilson_compute_into(data, period, v_factor, first, chosen, &mut out)?;
    Ok(TilsonOutput { values: out })
}

#[cfg(not(all(target_arch = "wasm32", feature = "wasm")))]
pub fn tilson_into(input: &TilsonInput, out: &mut [f64]) -> Result<(), TilsonError> {
    let (data, period, v_factor, first, len, chosen) = tilson_prepare(input, Kernel::Auto)?;

    if out.len() != len {
        return Err(TilsonError::OutputLengthMismatch {
            expected: len,
            got: out.len(),
        });
    }

    let warm = (first + 6 * (period - 1)).min(len);
    for i in 0..warm {
        out[i] = f64::from_bits(0x7ff8_0000_0000_0000);
    }

    tilson_compute_into(data, period, v_factor, first, chosen, out)?;
    Ok(())
}

#[inline]
pub fn tilson_into_slice(
    dst: &mut [f64],
    input: &TilsonInput,
    kern: Kernel,
) -> Result<(), TilsonError> {
    let (data, period, v_factor, first, _len, chosen) = tilson_prepare(input, kern)?;
    if dst.len() != data.len() {
        return Err(TilsonError::OutputLengthMismatch {
            expected: data.len(),
            got: dst.len(),
        });
    }
    tilson_compute_into(data, period, v_factor, first, chosen, dst)?;
    let warm = first + 6 * (period - 1);
    let warm_end = warm.min(dst.len());
    for v in &mut dst[..warm_end] {
        *v = f64::NAN;
    }
    Ok(())
}

#[inline]
pub fn tilson_scalar(
    data: &[f64],
    period: usize,
    v_factor: f64,
    first_valid: usize,
    out: &mut [f64],
) -> Result<(), TilsonError> {
    let len = data.len();
    let lookback_total = 6 * (period.saturating_sub(1));
    debug_assert_eq!(len, out.len());

    if len == 0 {
        return Err(TilsonError::EmptyInputData);
    }
    if period == 0 || len.saturating_sub(first_valid) < period {
        return Err(TilsonError::InvalidPeriod {
            period,
            data_len: len,
        });
    }
    if v_factor.is_nan() || v_factor.is_infinite() {
        return Err(TilsonError::InvalidVolumeFactor { v_factor });
    }
    if lookback_total + first_valid >= len {
        return Err(TilsonError::NotEnoughValidData {
            needed: lookback_total + 1,
            valid: len - first_valid,
        });
    }

    let k = 2.0 / (period as f64 + 1.0);
    let omk = 1.0 - k;
    let inv_p = 1.0 / (period as f64);

    let t = v_factor * v_factor;
    let c1 = -(t * v_factor);
    let c2 = 3.0 * (t - c1);
    let c3 = -6.0 * t - 3.0 * (v_factor - c1);
    let c4 = 1.0 + 3.0 * v_factor - c1 + 3.0 * t;

    let dp = unsafe { data.as_ptr().add(first_valid) };
    let outp = out.as_mut_ptr();

    let (mut e1, mut e2, mut e3, mut e4, mut e5, mut e6);

    let mut today = 0usize;

    let mut sum = 0.0;
    unsafe {
        let mut i = 0usize;
        while i + 4 <= period {
            let base = dp.add(today + i);
            sum += *base + *base.add(1) + *base.add(2) + *base.add(3);
            i += 4;
        }
        while i < period {
            sum += *dp.add(today + i);
            i += 1;
        }
    }
    e1 = sum * inv_p;
    today += period;

    let mut acc = e1;
    unsafe {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            acc += e1;
            today += 1;
            j += 1;
        }
    }
    e2 = acc * inv_p;

    acc = e2;
    unsafe {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            e2 = k * e1 + omk * e2;
            acc += e2;
            today += 1;
            j += 1;
        }
    }
    e3 = acc * inv_p;

    acc = e3;
    unsafe {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            e2 = k * e1 + omk * e2;
            e3 = k * e2 + omk * e3;
            acc += e3;
            today += 1;
            j += 1;
        }
    }
    e4 = acc * inv_p;

    acc = e4;
    unsafe {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            e2 = k * e1 + omk * e2;
            e3 = k * e2 + omk * e3;
            e4 = k * e3 + omk * e4;
            acc += e4;
            today += 1;
            j += 1;
        }
    }
    e5 = acc * inv_p;

    acc = e5;
    unsafe {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            e2 = k * e1 + omk * e2;
            e3 = k * e2 + omk * e3;
            e4 = k * e3 + omk * e4;
            e5 = k * e4 + omk * e5;
            acc += e5;
            today += 1;
            j += 1;
        }
    }
    e6 = acc * inv_p;

    let start_idx = first_valid + lookback_total;

    unsafe {
        *outp.add(start_idx) = c1 * e6 + c2 * e5 + c3 * e4 + c4 * e3;

        let mut dp_cur = dp.add(today);
        let dp_end = dp.add(len - first_valid);
        let mut out_cur = outp.add(start_idx + 1);
        while dp_cur < dp_end {
            let x = *dp_cur;
            e1 = k * x + omk * e1;
            e2 = k * e1 + omk * e2;
            e3 = k * e2 + omk * e3;
            e4 = k * e3 + omk * e4;
            e5 = k * e4 + omk * e5;
            e6 = k * e5 + omk * e6;

            *out_cur = c1 * e6 + c2 * e5 + c3 * e4 + c4 * e3;

            dp_cur = dp_cur.add(1);
            out_cur = out_cur.add(1);
        }
    }

    Ok(())
}

#[cfg(all(target_arch = "wasm32", target_feature = "simd128"))]
#[inline]
unsafe fn tilson_simd128(
    data: &[f64],
    period: usize,
    v_factor: f64,
    first_valid: usize,
    out: &mut [f64],
) -> Result<(), TilsonError> {
    use core::arch::wasm32::*;

    tilson_scalar(data, period, v_factor, first_valid, out)
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[target_feature(enable = "avx512f")]
#[inline]
pub unsafe fn tilson_avx512(
    data: &[f64],
    period: usize,
    v_factor: f64,
    first_valid: usize,
    out: &mut [f64],
) -> Result<(), TilsonError> {
    use core::arch::x86_64::*;

    #[inline(always)]
    unsafe fn dot4_avx512(
        e3: f64,
        e4: f64,
        e5: f64,
        e6: f64,
        c4: f64,
        c3: f64,
        c2: f64,
        c1: f64,
    ) -> f64 {
        let ve = _mm512_setr_pd(e3, e4, e5, e6, 0.0, 0.0, 0.0, 0.0);
        let vc = _mm512_setr_pd(c4, c3, c2, c1, 0.0, 0.0, 0.0, 0.0);
        let prod = _mm512_mul_pd(ve, vc);
        _mm512_reduce_add_pd(prod)
    }

    let len = data.len();
    let lookback_total = 6 * (period.saturating_sub(1));
    if len == 0 {
        return Err(TilsonError::EmptyInputData);
    }
    if period == 0 || len.saturating_sub(first_valid) < period {
        return Err(TilsonError::InvalidPeriod {
            period,
            data_len: len,
        });
    }
    if v_factor.is_nan() || v_factor.is_infinite() {
        return Err(TilsonError::InvalidVolumeFactor { v_factor });
    }
    if lookback_total + first_valid >= len {
        return Err(TilsonError::NotEnoughValidData {
            needed: lookback_total + 1,
            valid: len - first_valid,
        });
    }

    let k = 2.0 / (period as f64 + 1.0);
    let omk = 1.0 - k;
    let inv_p = 1.0 / (period as f64);

    let t = v_factor * v_factor;
    let c1 = -(t * v_factor);
    let c2 = 3.0 * (t - c1);
    let c3 = -6.0 * t - 3.0 * (v_factor - c1);
    let c4 = 1.0 + 3.0 * v_factor - c1 + 3.0 * t;

    let dp = data.as_ptr().add(first_valid);
    let outp = out.as_mut_ptr();

    let mut today = 0usize;
    let (mut e1, mut e2, mut e3, mut e4, mut e5, mut e6);

    let mut sum = 0.0;
    {
        let mut i = 0usize;
        while i + 4 <= period {
            let base = dp.add(today + i);
            sum += *base + *base.add(1) + *base.add(2) + *base.add(3);
            i += 4;
        }
        while i < period {
            sum += *dp.add(today + i);
            i += 1;
        }
    }
    e1 = sum * inv_p;
    today += period;

    let mut acc = e1;
    {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            acc += e1;
            today += 1;
            j += 1;
        }
    }
    e2 = acc * inv_p;

    acc = e2;
    {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            e2 = k * e1 + omk * e2;
            acc += e2;
            today += 1;
            j += 1;
        }
    }
    e3 = acc * inv_p;

    acc = e3;
    {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            e2 = k * e1 + omk * e2;
            e3 = k * e2 + omk * e3;
            acc += e3;
            today += 1;
            j += 1;
        }
    }
    e4 = acc * inv_p;

    acc = e4;
    {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            e2 = k * e1 + omk * e2;
            e3 = k * e2 + omk * e3;
            e4 = k * e3 + omk * e4;
            acc += e4;
            today += 1;
            j += 1;
        }
    }
    e5 = acc * inv_p;

    acc = e5;
    {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            e2 = k * e1 + omk * e2;
            e3 = k * e2 + omk * e3;
            e4 = k * e3 + omk * e4;
            e5 = k * e4 + omk * e5;
            acc += e5;
            today += 1;
            j += 1;
        }
    }
    e6 = acc * inv_p;

    let start_idx = first_valid + lookback_total;
    let end_idx = len - 1;

    *outp.add(start_idx) = dot4_avx512(e3, e4, e5, e6, c4, c3, c2, c1);

    let mut idx = start_idx + 1;
    while (first_valid + today) <= end_idx {
        let x = *dp.add(today);
        e1 = k * x + omk * e1;
        e2 = k * e1 + omk * e2;
        e3 = k * e2 + omk * e3;
        e4 = k * e3 + omk * e4;
        e5 = k * e4 + omk * e5;
        e6 = k * e5 + omk * e6;

        *outp.add(idx) = dot4_avx512(e3, e4, e5, e6, c4, c3, c2, c1);

        today += 1;
        idx += 1;
    }

    Ok(())
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[target_feature(enable = "avx2")]
#[inline]
pub unsafe fn tilson_avx2(
    data: &[f64],
    period: usize,
    v_factor: f64,
    first_valid: usize,
    out: &mut [f64],
) -> Result<(), TilsonError> {
    use core::arch::x86_64::*;

    #[inline(always)]
    unsafe fn dot4_avx2(
        e3: f64,
        e4: f64,
        e5: f64,
        e6: f64,
        c4: f64,
        c3: f64,
        c2: f64,
        c1: f64,
    ) -> f64 {
        let ve = _mm256_setr_pd(e3, e4, e5, e6);
        let vc = _mm256_setr_pd(c4, c3, c2, c1);
        let prod = _mm256_mul_pd(ve, vc);
        let lo = _mm256_castpd256_pd128(prod);
        let hi = _mm256_extractf128_pd(prod, 1);
        let s0 = _mm_hadd_pd(lo, lo);
        let s1 = _mm_hadd_pd(hi, hi);
        let sum = _mm_add_sd(s0, s1);
        _mm_cvtsd_f64(sum)
    }

    let len = data.len();
    let lookback_total = 6 * (period.saturating_sub(1));
    if len == 0 {
        return Err(TilsonError::EmptyInputData);
    }
    if period == 0 || len.saturating_sub(first_valid) < period {
        return Err(TilsonError::InvalidPeriod {
            period,
            data_len: len,
        });
    }
    if v_factor.is_nan() || v_factor.is_infinite() {
        return Err(TilsonError::InvalidVolumeFactor { v_factor });
    }
    if lookback_total + first_valid >= len {
        return Err(TilsonError::NotEnoughValidData {
            needed: lookback_total + 1,
            valid: len - first_valid,
        });
    }

    let k = 2.0 / (period as f64 + 1.0);
    let omk = 1.0 - k;
    let inv_p = 1.0 / (period as f64);

    let t = v_factor * v_factor;
    let c1 = -(t * v_factor);
    let c2 = 3.0 * (t - c1);
    let c3 = -6.0 * t - 3.0 * (v_factor - c1);
    let c4 = 1.0 + 3.0 * v_factor - c1 + 3.0 * t;

    let dp = data.as_ptr().add(first_valid);
    let outp = out.as_mut_ptr();

    let mut today = 0usize;
    let (mut e1, mut e2, mut e3, mut e4, mut e5, mut e6);

    let mut sum = 0.0;
    {
        let mut i = 0usize;
        while i + 4 <= period {
            let base = dp.add(today + i);
            sum += *base + *base.add(1) + *base.add(2) + *base.add(3);
            i += 4;
        }
        while i < period {
            sum += *dp.add(today + i);
            i += 1;
        }
    }
    e1 = sum * inv_p;
    today += period;

    let mut acc = e1;
    {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            acc += e1;
            today += 1;
            j += 1;
        }
    }
    e2 = acc * inv_p;

    acc = e2;
    {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            e2 = k * e1 + omk * e2;
            acc += e2;
            today += 1;
            j += 1;
        }
    }
    e3 = acc * inv_p;

    acc = e3;
    {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            e2 = k * e1 + omk * e2;
            e3 = k * e2 + omk * e3;
            acc += e3;
            today += 1;
            j += 1;
        }
    }
    e4 = acc * inv_p;

    acc = e4;
    {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            e2 = k * e1 + omk * e2;
            e3 = k * e2 + omk * e3;
            e4 = k * e3 + omk * e4;
            acc += e4;
            today += 1;
            j += 1;
        }
    }
    e5 = acc * inv_p;

    acc = e5;
    {
        let mut j = 1usize;
        while j < period {
            let x = *dp.add(today);
            e1 = k * x + omk * e1;
            e2 = k * e1 + omk * e2;
            e3 = k * e2 + omk * e3;
            e4 = k * e3 + omk * e4;
            e5 = k * e4 + omk * e5;
            acc += e5;
            today += 1;
            j += 1;
        }
    }
    e6 = acc * inv_p;

    let start_idx = first_valid + lookback_total;
    let end_idx = len - 1;

    *outp.add(start_idx) = dot4_avx2(e3, e4, e5, e6, c4, c3, c2, c1);

    let mut idx = start_idx + 1;
    while (first_valid + today) <= end_idx {
        let x = *dp.add(today);
        e1 = k * x + omk * e1;
        e2 = k * e1 + omk * e2;
        e3 = k * e2 + omk * e3;
        e4 = k * e3 + omk * e4;
        e5 = k * e4 + omk * e5;
        e6 = k * e5 + omk * e6;

        *outp.add(idx) = dot4_avx2(e3, e4, e5, e6, c4, c3, c2, c1);

        today += 1;
        idx += 1;
    }

    Ok(())
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline]
pub unsafe fn tilson_avx512_short(
    data: &[f64],
    period: usize,
    v_factor: f64,
    first_valid: usize,
    out: &mut [f64],
) -> Result<(), TilsonError> {
    tilson_avx512(data, period, v_factor, first_valid, out)
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline]
pub unsafe fn tilson_avx512_long(
    data: &[f64],
    period: usize,
    v_factor: f64,
    first_valid: usize,
    out: &mut [f64],
) -> Result<(), TilsonError> {
    tilson_avx512(data, period, v_factor, first_valid, out)
}

#[inline]
pub fn tilson_batch_with_kernel(
    data: &[f64],
    sweep: &TilsonBatchRange,
    k: Kernel,
) -> Result<TilsonBatchOutput, TilsonError> {
    let kernel = match k {
        Kernel::Auto => detect_best_batch_kernel(),
        other if other.is_batch() => other,
        _ => return Err(TilsonError::InvalidKernelForBatch(k)),
    };

    let simd = match kernel {
        Kernel::Avx512Batch => Kernel::Avx512,
        Kernel::Avx2Batch => Kernel::Avx2,
        Kernel::ScalarBatch => Kernel::Scalar,
        _ => unreachable!(),
    };
    tilson_batch_par_slice(data, sweep, simd)
}

#[derive(Clone, Debug)]
pub struct TilsonBatchRange {
    pub period: (usize, usize, usize),
    pub volume_factor: (f64, f64, f64),
}

impl Default for TilsonBatchRange {
    fn default() -> Self {
        Self {
            period: (5, 254, 1),
            volume_factor: (0.0, 0.0, 0.0),
        }
    }
}

#[derive(Clone, Debug, Default)]
pub struct TilsonBatchBuilder {
    range: TilsonBatchRange,
    kernel: Kernel,
}

impl TilsonBatchBuilder {
    pub fn new() -> Self {
        Self::default()
    }
    pub fn kernel(mut self, k: Kernel) -> Self {
        self.kernel = k;
        self
    }

    #[inline]
    pub fn period_range(mut self, start: usize, end: usize, step: usize) -> Self {
        self.range.period = (start, end, step);
        self
    }
    #[inline]
    pub fn period_static(mut self, p: usize) -> Self {
        self.range.period = (p, p, 0);
        self
    }

    #[inline]
    pub fn volume_factor_range(mut self, start: f64, end: f64, step: f64) -> Self {
        self.range.volume_factor = (start, end, step);
        self
    }
    #[inline]
    pub fn volume_factor_static(mut self, v: f64) -> Self {
        self.range.volume_factor = (v, v, 0.0);
        self
    }

    pub fn apply_slice(self, data: &[f64]) -> Result<TilsonBatchOutput, TilsonError> {
        tilson_batch_with_kernel(data, &self.range, self.kernel)
    }
    pub fn with_default_slice(data: &[f64], k: Kernel) -> Result<TilsonBatchOutput, TilsonError> {
        TilsonBatchBuilder::new().kernel(k).apply_slice(data)
    }

    pub fn apply_candles(self, c: &Candles, src: &str) -> Result<TilsonBatchOutput, TilsonError> {
        let slice = source_type(c, src);
        self.apply_slice(slice)
    }

    pub fn with_default_candles(c: &Candles) -> Result<TilsonBatchOutput, TilsonError> {
        TilsonBatchBuilder::new()
            .kernel(Kernel::Auto)
            .apply_candles(c, "close")
    }
}

#[derive(Clone, Debug)]
pub struct TilsonBatchOutput {
    pub values: Vec<f64>,
    pub combos: Vec<TilsonParams>,
    pub rows: usize,
    pub cols: usize,
}
impl TilsonBatchOutput {
    pub fn row_for_params(&self, p: &TilsonParams) -> Option<usize> {
        self.combos.iter().position(|c| {
            c.period.unwrap_or(5) == p.period.unwrap_or(5)
                && (c.volume_factor.unwrap_or(0.0) - p.volume_factor.unwrap_or(0.0)).abs() < 1e-12
        })
    }
    pub fn values_for(&self, p: &TilsonParams) -> Option<&[f64]> {
        self.row_for_params(p).map(|row| {
            let start = row * self.cols;
            &self.values[start..start + self.cols]
        })
    }
}

#[inline(always)]
fn expand_grid(r: &TilsonBatchRange) -> Result<Vec<TilsonParams>, TilsonError> {
    fn axis_usize((start, end, step): (usize, usize, usize)) -> Result<Vec<usize>, TilsonError> {
        if step == 0 || start == end {
            return Ok(vec![start]);
        }
        if start < end {
            return Ok((start..=end).step_by(step).collect());
        }

        let mut v = Vec::new();
        let mut cur = start;
        loop {
            v.push(cur);
            match cur.checked_sub(step) {
                Some(next) if next >= end => {
                    cur = next;
                }
                _ => break,
            }
        }
        if v.is_empty() {
            Err(TilsonError::InvalidRangeUsize { start, end, step })
        } else {
            Ok(v)
        }
    }
    fn axis_f64((start, end, step): (f64, f64, f64)) -> Result<Vec<f64>, TilsonError> {
        if step.abs() < 1e-12 || (start - end).abs() < 1e-12 {
            return Ok(vec![start]);
        }
        let mut v = Vec::new();
        let mut x = start;
        if step > 0.0 {
            while x <= end + 1e-12 {
                v.push(x);
                x += step;
            }
        } else {
            while x >= end - 1e-12 {
                v.push(x);
                x += step;
            }
        }
        if v.is_empty() {
            Err(TilsonError::InvalidRangeF64 { start, end, step })
        } else {
            Ok(v)
        }
    }

    let periods = axis_usize(r.period)?;
    let v_factors = axis_f64(r.volume_factor)?;

    let mut out = Vec::with_capacity(periods.len().saturating_mul(v_factors.len()));
    for &p in &periods {
        for &v in &v_factors {
            out.push(TilsonParams {
                period: Some(p),
                volume_factor: Some(v),
            });
        }
    }
    if out.is_empty() {
        return Err(TilsonError::InvalidInput("empty parameter sweep"));
    }
    Ok(out)
}

#[inline(always)]
pub fn tilson_batch_slice(
    data: &[f64],
    sweep: &TilsonBatchRange,
    kern: Kernel,
) -> Result<TilsonBatchOutput, TilsonError> {
    tilson_batch_inner(data, sweep, kern, false)
}

#[inline(always)]
pub fn tilson_batch_par_slice(
    data: &[f64],
    sweep: &TilsonBatchRange,
    kern: Kernel,
) -> Result<TilsonBatchOutput, TilsonError> {
    tilson_batch_inner(data, sweep, kern, true)
}

#[inline(always)]
fn tilson_batch_inner(
    data: &[f64],
    sweep: &TilsonBatchRange,
    kern: Kernel,
    parallel: bool,
) -> Result<TilsonBatchOutput, TilsonError> {
    let combos = expand_grid(sweep)?;

    if data.is_empty() {
        return Err(TilsonError::EmptyInputData);
    }

    let first = data
        .iter()
        .position(|x| !x.is_nan())
        .ok_or(TilsonError::AllValuesNaN)?;
    let max_p = combos.iter().map(|c| c.period.unwrap()).max().unwrap();
    if data.len() - first < 6 * (max_p - 1) + 1 {
        return Err(TilsonError::NotEnoughValidData {
            needed: 6 * (max_p - 1) + 1,
            valid: data.len() - first,
        });
    }
    let rows = combos.len();
    let cols = data.len();
    let warm: Vec<usize> = combos
        .iter()
        .map(|c| first + 6 * (c.period.unwrap() - 1))
        .collect();

    let mut raw = make_uninit_matrix(rows, cols);
    unsafe { init_matrix_prefixes(&mut raw, cols, &warm) };

    let do_row = |row: usize, dst_mu: &mut [MaybeUninit<f64>]| unsafe {
        let period = combos[row].period.unwrap();
        let v_factor = combos[row].volume_factor.unwrap();

        let out_row =
            core::slice::from_raw_parts_mut(dst_mu.as_mut_ptr() as *mut f64, dst_mu.len());
        tilson_row_scalar(data, first, period, v_factor, out_row);
    };

    if parallel {
        #[cfg(not(target_arch = "wasm32"))]
        {
            raw.par_chunks_mut(cols)
                .enumerate()
                .for_each(|(row, slice)| do_row(row, slice));
        }

        #[cfg(target_arch = "wasm32")]
        {
            for (row, slice) in raw.chunks_mut(cols).enumerate() {
                do_row(row, slice);
            }
        }
    } else {
        for (row, slice) in raw.chunks_mut(cols).enumerate() {
            do_row(row, slice);
        }
    }

    let mut guard = core::mem::ManuallyDrop::new(raw);
    let total_cells = rows
        .checked_mul(cols)
        .ok_or(TilsonError::InvalidInput("rows*cols overflow"))?;
    let values = unsafe {
        Vec::from_raw_parts(
            guard.as_mut_ptr() as *mut f64,
            total_cells,
            guard.capacity(),
        )
    };

    Ok(TilsonBatchOutput {
        values,
        combos,
        rows,
        cols,
    })
}

#[inline(always)]
pub unsafe fn tilson_row_scalar(
    data: &[f64],
    first: usize,
    period: usize,
    v_factor: f64,
    out: &mut [f64],
) {
    let _ = tilson_scalar(data, period, v_factor, first, out);
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn tilson_row_avx2(
    data: &[f64],
    first: usize,
    period: usize,
    v_factor: f64,
    out: &mut [f64],
) {
    tilson_row_scalar(data, first, period, v_factor, out);
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn tilson_row_avx512(
    data: &[f64],
    first: usize,
    period: usize,
    v_factor: f64,
    out: &mut [f64],
) {
    tilson_row_scalar(data, first, period, v_factor, out);
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn tilson_row_avx512_short(
    data: &[f64],
    first: usize,
    period: usize,
    v_factor: f64,
    out: &mut [f64],
) {
    tilson_row_scalar(data, first, period, v_factor, out);
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn tilson_row_avx512_long(
    data: &[f64],
    first: usize,
    period: usize,
    v_factor: f64,
    out: &mut [f64],
) {
    tilson_row_scalar(data, first, period, v_factor, out);
}

#[derive(Debug, Clone)]
pub struct TilsonStream {
    period: usize,
    v_factor: f64,

    e1: f64,
    e2: f64,
    e3: f64,
    e4: f64,
    e5: f64,
    e6: f64,

    k: f64,
    one_minus_k: f64,
    inv_p: f64,
    c1: f64,
    c2: f64,
    c3: f64,
    c4: f64,

    phase: u8,
    in_phase_count: usize,
    sum_e1: f64,
    acc: f64,

    lookback_total: usize,
    values_seen: usize,
}

impl TilsonStream {
    pub fn try_new(params: TilsonParams) -> Result<Self, TilsonError> {
        let period = params.period.unwrap_or(5);
        let v_factor = params.volume_factor.unwrap_or(0.0);

        if period == 0 {
            return Err(TilsonError::InvalidPeriod {
                period,
                data_len: 0,
            });
        }
        if v_factor.is_nan() || v_factor.is_infinite() {
            return Err(TilsonError::InvalidVolumeFactor { v_factor });
        }

        let k = 2.0 / (period as f64 + 1.0);
        let one_minus_k = 1.0 - k;
        let inv_p = 1.0 / (period as f64);

        let t = v_factor * v_factor;
        let c1 = -(t * v_factor);
        let c2 = 3.0 * (t - c1);
        let c3 = -6.0 * t - 3.0 * (v_factor - c1);
        let c4 = 1.0 + 3.0 * v_factor - c1 + 3.0 * t;

        Ok(Self {
            period,
            v_factor,
            e1: 0.0,
            e2: 0.0,
            e3: 0.0,
            e4: 0.0,
            e5: 0.0,
            e6: 0.0,
            k,
            one_minus_k,
            inv_p,
            c1,
            c2,
            c3,
            c4,
            phase: 0,
            in_phase_count: 0,
            sum_e1: 0.0,
            acc: 0.0,
            lookback_total: 6 * (period - 1),
            values_seen: 0,
        })
    }

    #[inline(always)]
    pub fn update(&mut self, value: f64) -> Option<f64> {
        self.values_seen = self.values_seen.wrapping_add(1);

        match self.phase {
            0 => {
                self.sum_e1 += value;
                self.in_phase_count += 1;
                if self.in_phase_count == self.period {
                    self.e1 = self.sum_e1 * self.inv_p;
                    self.in_phase_count = 0;

                    if self.period == 1 {
                        self.e2 = self.e1;
                        self.e3 = self.e2;
                        self.e4 = self.e3;
                        self.e5 = self.e4;
                        self.e6 = self.e5;
                        self.phase = 6;
                        return Some(self.combine_exact());
                    }

                    self.acc = self.e1;
                    self.phase = 1;
                }
                None
            }

            1 => {
                self.cascade_update_upto(value, 1);
                self.acc += self.e1;
                self.in_phase_count += 1;
                if self.in_phase_count == self.period - 1 {
                    self.e2 = self.acc * self.inv_p;
                    self.acc = self.e2;
                    self.in_phase_count = 0;
                    self.phase = 2;
                }
                None
            }

            2 => {
                self.cascade_update_upto(value, 2);
                self.acc += self.e2;
                self.in_phase_count += 1;
                if self.in_phase_count == self.period - 1 {
                    self.e3 = self.acc * self.inv_p;
                    self.acc = self.e3;
                    self.in_phase_count = 0;
                    self.phase = 3;
                }
                None
            }

            3 => {
                self.cascade_update_upto(value, 3);
                self.acc += self.e3;
                self.in_phase_count += 1;
                if self.in_phase_count == self.period - 1 {
                    self.e4 = self.acc * self.inv_p;
                    self.acc = self.e4;
                    self.in_phase_count = 0;
                    self.phase = 4;
                }
                None
            }

            4 => {
                self.cascade_update_upto(value, 4);
                self.acc += self.e4;
                self.in_phase_count += 1;
                if self.in_phase_count == self.period - 1 {
                    self.e5 = self.acc * self.inv_p;
                    self.acc = self.e5;
                    self.in_phase_count = 0;
                    self.phase = 5;
                }
                None
            }

            5 => {
                self.cascade_update_upto(value, 5);
                self.acc += self.e5;
                self.in_phase_count += 1;
                if self.in_phase_count == self.period - 1 {
                    self.e6 = self.acc * self.inv_p;
                    self.phase = 6;
                    self.in_phase_count = 0;

                    return Some(self.combine_exact());
                }
                None
            }

            _ => {
                self.cascade_update_upto(value, 6);
                Some(self.combine_exact())
            }
        }
    }

    #[inline(always)]
    fn cascade_update_upto(&mut self, x: f64, upto: u8) {
        let k = self.k;
        let omk = self.one_minus_k;

        self.e1 = k * x + omk * self.e1;
        if upto == 1 {
            return;
        }

        self.e2 = k * self.e1 + omk * self.e2;
        if upto == 2 {
            return;
        }

        self.e3 = k * self.e2 + omk * self.e3;
        if upto == 3 {
            return;
        }

        self.e4 = k * self.e3 + omk * self.e4;
        if upto == 4 {
            return;
        }

        self.e5 = k * self.e4 + omk * self.e5;
        if upto == 5 {
            return;
        }

        self.e6 = k * self.e5 + omk * self.e6;
    }

    #[inline(always)]
    fn combine_exact(&self) -> f64 {
        let s0 = self.c1 * self.e6 + self.c2 * self.e5;
        let s1 = s0 + self.c3 * self.e4;
        s1 + self.c4 * self.e3
    }
}

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

    fn check_tilson_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 = TilsonParams {
            period: None,
            volume_factor: None,
        };
        let input = TilsonInput::from_candles(&candles, "close", default_params);
        let output = tilson_with_kernel(&input, kernel)?;
        assert_eq!(output.values.len(), candles.close.len());
        Ok(())
    }

    fn check_tilson_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 = TilsonInput::from_candles(&candles, "close", TilsonParams::default());
        let result = tilson_with_kernel(&input, kernel)?;
        let expected_last_five = [
            59304.716332473254,
            59283.56868015526,
            59261.16173577631,
            59240.25895948583,
            59203.544843167765,
        ];
        let start = result.values.len().saturating_sub(5);
        for (i, &val) in result.values[start..].iter().enumerate() {
            let diff = (val - expected_last_five[i]).abs();
            assert!(
                diff < 1e-8,
                "[{}] TILSON {:?} mismatch at idx {}: got {}, expected {}",
                test_name,
                kernel,
                i,
                val,
                expected_last_five[i]
            );
        }
        Ok(())
    }

    fn check_tilson_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 = TilsonInput::with_default_candles(&candles);
        match input.data {
            TilsonData::Candles { source, .. } => assert_eq!(source, "close"),
            _ => panic!("Expected TilsonData::Candles"),
        }
        let output = tilson_with_kernel(&input, kernel)?;
        assert_eq!(output.values.len(), candles.close.len());
        Ok(())
    }

    fn check_tilson_zero_period(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let input_data = [10.0, 20.0, 30.0];
        let params = TilsonParams {
            period: Some(0),
            volume_factor: None,
        };
        let input = TilsonInput::from_slice(&input_data, params);
        let res = tilson_with_kernel(&input, kernel);
        assert!(
            res.is_err(),
            "[{}] TILSON should fail with zero period",
            test_name
        );
        Ok(())
    }

    fn check_tilson_empty_input(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let input_data: [f64; 0] = [];
        let params = TilsonParams {
            period: Some(5),
            volume_factor: Some(0.0),
        };
        let input = TilsonInput::from_slice(&input_data, params);
        let res = tilson_with_kernel(&input, kernel);
        assert!(
            res.is_err(),
            "[{}] TILSON should fail with empty input",
            test_name
        );
        if let Err(e) = res {
            assert!(
                matches!(e, TilsonError::EmptyInputData),
                "[{}] Expected EmptyInputData error",
                test_name
            );
        }
        Ok(())
    }

    fn check_tilson_period_exceeds_length(
        test_name: &str,
        kernel: Kernel,
    ) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let data_small = [10.0, 20.0, 30.0];
        let params = TilsonParams {
            period: Some(10),
            volume_factor: None,
        };
        let input = TilsonInput::from_slice(&data_small, params);
        let res = tilson_with_kernel(&input, kernel);
        assert!(
            res.is_err(),
            "[{}] TILSON should fail with period exceeding length",
            test_name
        );
        Ok(())
    }

    fn check_tilson_very_small_dataset(
        test_name: &str,
        kernel: Kernel,
    ) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let single_point = [42.0];
        let params = TilsonParams {
            period: Some(9),
            volume_factor: None,
        };
        let input = TilsonInput::from_slice(&single_point, params);
        let res = tilson_with_kernel(&input, kernel);
        assert!(
            res.is_err(),
            "[{}] TILSON should fail with insufficient data",
            test_name
        );
        Ok(())
    }

    fn check_tilson_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 first_params = TilsonParams {
            period: Some(5),
            volume_factor: None,
        };
        let first_input = TilsonInput::from_candles(&candles, "close", first_params);
        let first_result = tilson_with_kernel(&first_input, kernel)?;

        let second_params = TilsonParams {
            period: Some(3),
            volume_factor: Some(0.7),
        };
        let second_input = TilsonInput::from_slice(&first_result.values, second_params);
        let second_result = tilson_with_kernel(&second_input, kernel)?;

        assert_eq!(second_result.values.len(), first_result.values.len());
        for i in 240..second_result.values.len() {
            assert!(second_result.values[i].is_finite());
        }
        Ok(())
    }

    fn check_tilson_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 = TilsonInput::from_candles(
            &candles,
            "close",
            TilsonParams {
                period: Some(5),
                volume_factor: Some(0.0),
            },
        );
        let res = tilson_with_kernel(&input, kernel)?;
        assert_eq!(res.values.len(), candles.close.len());
        if res.values.len() > 240 {
            for (i, &val) in res.values[240..].iter().enumerate() {
                assert!(
                    !val.is_nan(),
                    "[{}] Found unexpected NaN at out-index {}",
                    test_name,
                    240 + i
                );
            }
        }
        Ok(())
    }

    fn check_tilson_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 period = 5;
        let v_factor = 0.0;

        let input = TilsonInput::from_candles(
            &candles,
            "close",
            TilsonParams {
                period: Some(period),
                volume_factor: Some(v_factor),
            },
        );
        let batch_output = tilson_with_kernel(&input, kernel)?.values;

        let mut stream = TilsonStream::try_new(TilsonParams {
            period: Some(period),
            volume_factor: Some(v_factor),
        })?;

        let mut stream_values = Vec::with_capacity(candles.close.len());
        for &price in &candles.close {
            match stream.update(price) {
                Some(val) => stream_values.push(val),
                None => stream_values.push(f64::NAN),
            }
        }
        assert_eq!(batch_output.len(), stream_values.len());
        for (i, (&b, &s)) in batch_output.iter().zip(stream_values.iter()).enumerate() {
            if b.is_nan() && s.is_nan() {
                continue;
            }
            let diff = (b - s).abs();
            assert!(
                diff < 1e-9,
                "[{}] TILSON streaming f64 mismatch at idx {}: batch={}, stream={}, diff={}",
                test_name,
                i,
                b,
                s,
                diff
            );
        }
        Ok(())
    }

    macro_rules! generate_all_tilson_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);
                    }
                )*
            }
        }
    }

    #[cfg(debug_assertions)]
    fn check_tilson_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 test_periods = vec![3, 5, 8, 10, 15, 20, 30, 50];
        let test_v_factors = vec![0.0, 0.1, 0.3, 0.5, 0.7, 0.9, 1.0];

        for &period in &test_periods {
            for &v_factor in &test_v_factors {
                let params = TilsonParams {
                    period: Some(period),
                    volume_factor: Some(v_factor),
                };
                let input = TilsonInput::from_candles(&candles, "close", params);

                if candles.close.len() < 6 * (period - 1) + 1 {
                    continue;
                }

                let output = match tilson_with_kernel(&input, kernel) {
                    Ok(o) => o,
                    Err(_) => continue,
                };

                for (i, &val) in output.values.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 {} with period {} and v_factor {}",
                            test_name, val, bits, i, period, v_factor
                        );
                    }

                    if bits == 0x22222222_22222222 {
                        panic!(
                            "[{}] Found init_matrix_prefixes poison value {} (0x{:016X}) at index {} with period {} and v_factor {}",
                            test_name, val, bits, i, period, v_factor
                        );
                    }

                    if bits == 0x33333333_33333333 {
                        panic!(
                            "[{}] Found make_uninit_matrix poison value {} (0x{:016X}) at index {} with period {} and v_factor {}",
                            test_name, val, bits, i, period, v_factor
                        );
                    }
                }
            }
        }

        Ok(())
    }

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

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

        let strat = (1usize..=30).prop_flat_map(|period| {
            let min_len = (6 * period.saturating_sub(1) + 1).max(period);
            (
                prop::collection::vec(
                    (-1e6f64..1e6f64).prop_filter("finite", |x| x.is_finite()),
                    min_len..400,
                ),
                Just(period),
                0.0f64..=1.0f64,
            )
        });

        proptest::test_runner::TestRunner::default()
            .run(&strat, |(data, period, volume_factor)| {
                let params = TilsonParams {
                    period: Some(period),
                    volume_factor: Some(volume_factor),
                };
                let input = TilsonInput::from_slice(&data, params);

                let TilsonOutput { values: out } = tilson_with_kernel(&input, kernel).unwrap();

                let TilsonOutput { values: ref_out } =
                    tilson_with_kernel(&input, Kernel::Scalar).unwrap();

                prop_assert_eq!(out.len(), data.len(), "Output length mismatch");

                let warmup_end = 6 * (period - 1);

                for i in 0..warmup_end.min(out.len()) {
                    prop_assert!(
                        out[i].is_nan(),
                        "Expected NaN during warmup at index {}, got {}",
                        i,
                        out[i]
                    );
                }

                let is_constant_data = data.windows(2).all(|w| (w[0] - w[1]).abs() < 1e-12);

                for i in warmup_end..data.len() {
                    let y = out[i];
                    let r = ref_out[i];

                    if y.is_finite() && r.is_finite() {
                        let y_bits = y.to_bits();
                        let r_bits = r.to_bits();
                        let ulp_diff = y_bits.abs_diff(r_bits);

                        prop_assert!(
                            (y - r).abs() <= 1e-9 || ulp_diff <= 8,
                            "SIMD mismatch at idx {}: {} vs {} (ULP={})",
                            i,
                            y,
                            r,
                            ulp_diff
                        );
                    } else {
                        prop_assert_eq!(
                            y.to_bits(),
                            r.to_bits(),
                            "Non-finite value mismatch at index {}",
                            i
                        );
                    }

                    if is_constant_data && i >= warmup_end + period {
                        let const_val = data[0];
                        prop_assert!(
                            (y - const_val).abs() <= 1e-9,
                            "Constant data property failed at idx {}: expected {}, got {}",
                            i,
                            const_val,
                            y
                        );
                    }
                }

                if period == 1 {
                    for i in 0..data.len() {
                        if out[i].is_finite() && data[i].is_finite() {
                            let tol = (data[i].abs() * 1e-10).max(1e-9);
                            prop_assert!(
                                (out[i] - data[i]).abs() <= tol,
                                "Period=1 property failed at idx {}: expected {}, got {}, diff={}",
                                i,
                                data[i],
                                out[i],
                                (out[i] - data[i]).abs()
                            );
                        }
                    }
                }

                if volume_factor == 0.0 && warmup_end < data.len() {
                    for i in warmup_end..data.len() {
                        prop_assert!(
                            out[i].is_finite(),
                            "With volume_factor=0, output should be finite at idx {}",
                            i
                        );
                    }
                }

                Ok(())
            })
            .unwrap();

        Ok(())
    }

    #[cfg(not(feature = "proptest"))]
    fn check_tilson_property(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        Ok(())
    }

    generate_all_tilson_tests!(
        check_tilson_partial_params,
        check_tilson_accuracy,
        check_tilson_default_candles,
        check_tilson_zero_period,
        check_tilson_empty_input,
        check_tilson_period_exceeds_length,
        check_tilson_very_small_dataset,
        check_tilson_reinput,
        check_tilson_nan_handling,
        check_tilson_streaming,
        check_tilson_no_poison,
        check_tilson_property
    );

    #[test]
    fn test_volume_factor_validation() {
        let data = vec![
            1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0,
        ];

        let params1 = TilsonParams {
            period: Some(3),
            volume_factor: Some(1.5),
        };
        let input1 = TilsonInput::from_slice(&data, params1);
        assert!(
            tilson(&input1).is_ok(),
            "volume_factor=1.5 should be accepted"
        );

        let params2 = TilsonParams {
            period: Some(3),
            volume_factor: Some(-0.5),
        };
        let input2 = TilsonInput::from_slice(&data, params2);
        assert!(
            tilson(&input2).is_ok(),
            "volume_factor=-0.5 should be accepted"
        );

        let params3 = TilsonParams {
            period: Some(3),
            volume_factor: Some(f64::NAN),
        };
        let input3 = TilsonInput::from_slice(&data, params3);
        assert!(
            tilson(&input3).is_err(),
            "volume_factor=NaN should be rejected"
        );

        let params4 = TilsonParams {
            period: Some(3),
            volume_factor: Some(f64::INFINITY),
        };
        let input4 = TilsonInput::from_slice(&data, params4);
        assert!(
            tilson(&input4).is_err(),
            "volume_factor=INFINITY should be rejected"
        );
    }

    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 = TilsonBatchBuilder::new()
            .kernel(kernel)
            .apply_candles(&c, "close")?;

        let def = TilsonParams::default();
        let row = output.values_for(&def).expect("default row missing");

        assert_eq!(row.len(), c.close.len());

        let expected = [
            59304.716332473254,
            59283.56868015526,
            59261.16173577631,
            59240.25895948583,
            59203.544843167765,
        ];
        let start = row.len() - 5;
        for (i, &v) in row[start..].iter().enumerate() {
            assert!(
                (v - expected[i]).abs() < 1e-8,
                "[{test}] default-row mismatch at idx {i}: {v} vs {expected:?}"
            );
        }
        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);
                }
            }
        };
    }

    #[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 test_configs = vec![
            (3, 10, 2, 0.0, 0.5, 0.25),
            (5, 20, 5, 0.0, 1.0, 0.2),
            (10, 50, 10, 0.3, 0.7, 0.2),
            (20, 40, 10, 0.0, 1.0, 0.5),
            (5, 5, 1, 0.0, 1.0, 0.1),
            (15, 15, 1, 0.5, 0.5, 0.1),
        ];

        for (p_start, p_end, p_step, v_start, v_end, v_step) in test_configs {
            let output = TilsonBatchBuilder::new()
                .kernel(kernel)
                .period_range(p_start, p_end, p_step)
                .volume_factor_range(v_start, v_end, v_step)
                .apply_candles(&c, "close")?;

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

                let bits = val.to_bits();
                let row = idx / output.cols;
                let col = idx % output.cols;
                let params = output.combos.get(row);
                let period = params.map(|p| p.period.unwrap_or(0)).unwrap_or(0);
                let v_factor = params
                    .map(|p| p.volume_factor.unwrap_or(0.0))
                    .unwrap_or(0.0);

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

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

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

        Ok(())
    }

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

    gen_batch_tests!(check_batch_default_row);
    gen_batch_tests!(check_batch_no_poison);

    #[test]
    fn test_tilson_into_matches_api() -> Result<(), Box<dyn Error>> {
        let mut data = Vec::with_capacity(256);
        data.extend_from_slice(&[f64::NAN, f64::NAN, f64::NAN, f64::NAN]);
        for i in 0..252u32 {
            data.push((i as f64).sin() * 100.0 + (i as f64) * 0.1);
        }

        let input = TilsonInput::from_slice(&data, TilsonParams::default());

        let baseline = tilson(&input)?.values;

        let mut out = vec![0.0; data.len()];

        #[cfg(not(all(target_arch = "wasm32", feature = "wasm")))]
        {
            tilson_into(&input, &mut out)?;

            assert_eq!(baseline.len(), out.len());
            for (a, b) in baseline.iter().zip(out.iter()) {
                let equal = (a.is_nan() && b.is_nan()) || (a == b);
                assert!(equal, "Mismatch: baseline={} into={}", a, b);
            }
        }

        Ok(())
    }
}

#[inline]
fn tilson_prepare<'a>(
    input: &'a TilsonInput,
    kernel: Kernel,
) -> Result<(&'a [f64], usize, f64, usize, usize, Kernel), TilsonError> {
    let data: &[f64] = input.as_ref();

    if data.is_empty() {
        return Err(TilsonError::EmptyInputData);
    }

    let first = data
        .iter()
        .position(|x| !x.is_nan())
        .ok_or(TilsonError::AllValuesNaN)?;
    let len = data.len();
    let period = input.get_period();
    let v_factor = input.get_volume_factor();

    if period == 0 || period > len {
        return Err(TilsonError::InvalidPeriod {
            period,
            data_len: len,
        });
    }
    if (len - first) < period {
        return Err(TilsonError::NotEnoughValidData {
            needed: period,
            valid: len - first,
        });
    }
    if v_factor.is_nan() || v_factor.is_infinite() {
        return Err(TilsonError::InvalidVolumeFactor { v_factor });
    }

    let lookback_total = 6 * (period - 1);
    if (len - first) < lookback_total + 1 {
        return Err(TilsonError::NotEnoughValidData {
            needed: lookback_total + 1,
            valid: len - first,
        });
    }

    let chosen = match kernel {
        Kernel::Auto => Kernel::Scalar,
        other => other,
    };

    Ok((data, period, v_factor, first, len, chosen))
}

#[inline]
fn tilson_compute_into(
    data: &[f64],
    period: usize,
    v_factor: f64,
    first: usize,
    chosen: Kernel,
    out: &mut [f64],
) -> Result<(), TilsonError> {
    unsafe {
        #[cfg(all(target_arch = "wasm32", target_feature = "simd128"))]
        {
            if matches!(chosen, Kernel::Scalar | Kernel::ScalarBatch) {
                tilson_simd128(data, period, v_factor, first, out)?;
                return Ok(());
            }
        }

        match chosen {
            Kernel::Scalar | Kernel::ScalarBatch => {
                tilson_scalar(data, period, v_factor, first, out)?
            }

            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx2 | Kernel::Avx2Batch => tilson_scalar(data, period, v_factor, first, out)?,
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx512 | Kernel::Avx512Batch => {
                tilson_scalar(data, period, v_factor, first, out)?
            }
            #[cfg(not(all(feature = "nightly-avx", target_arch = "x86_64")))]
            Kernel::Avx2 | Kernel::Avx2Batch | Kernel::Avx512 | Kernel::Avx512Batch => {
                tilson_scalar(data, period, v_factor, first, out)?
            }
            Kernel::Auto => unreachable!(),
        }
    }
    Ok(())
}

#[inline(always)]
fn tilson_batch_inner_into(
    data: &[f64],
    sweep: &TilsonBatchRange,
    kern: Kernel,
    parallel: bool,
    out: &mut [f64],
) -> Result<Vec<TilsonParams>, TilsonError> {
    let combos = expand_grid(sweep)?;

    if data.is_empty() {
        return Err(TilsonError::EmptyInputData);
    }

    let first = data
        .iter()
        .position(|x| !x.is_nan())
        .ok_or(TilsonError::AllValuesNaN)?;
    let max_p = combos.iter().map(|c| c.period.unwrap()).max().unwrap();

    if data.len() - first < 6 * (max_p - 1) + 1 {
        return Err(TilsonError::NotEnoughValidData {
            needed: 6 * (max_p - 1) + 1,
            valid: data.len() - first,
        });
    }

    let rows = combos.len();
    let cols = data.len();

    let expected = rows
        .checked_mul(cols)
        .ok_or(TilsonError::InvalidInput("rows*cols overflow"))?;
    if out.len() != expected {
        return Err(TilsonError::OutputLengthMismatch {
            expected,
            got: out.len(),
        });
    }

    let warm: Vec<usize> = combos
        .iter()
        .map(|c| (6 * (c.period.unwrap() - 1) + first).min(cols))
        .collect();

    let out_uninit = unsafe {
        std::slice::from_raw_parts_mut(out.as_mut_ptr() as *mut MaybeUninit<f64>, out.len())
    };

    unsafe { init_matrix_prefixes(out_uninit, cols, &warm) };

    let do_row = |row: usize, dst_mu: &mut [MaybeUninit<f64>]| unsafe {
        let period = combos[row].period.unwrap();
        let v_factor = combos[row].volume_factor.unwrap();

        let out_row =
            core::slice::from_raw_parts_mut(dst_mu.as_mut_ptr() as *mut f64, dst_mu.len());

        match kern {
            Kernel::Scalar | Kernel::ScalarBatch => {
                tilson_row_scalar(data, first, period, v_factor, out_row)
            }
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx2 | Kernel::Avx2Batch => {
                tilson_row_avx2(data, first, period, v_factor, out_row)
            }
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx512 | Kernel::Avx512Batch => {
                tilson_row_avx512(data, first, period, v_factor, out_row)
            }
            #[cfg(not(all(feature = "nightly-avx", target_arch = "x86_64")))]
            Kernel::Avx2 | Kernel::Avx512 | Kernel::Avx2Batch | Kernel::Avx512Batch => {
                tilson_row_scalar(data, first, period, v_factor, out_row)
            }
            _ => unreachable!(),
        }
    };

    if parallel {
        #[cfg(not(target_arch = "wasm32"))]
        {
            out_uninit
                .par_chunks_mut(cols)
                .enumerate()
                .for_each(|(row, slice)| do_row(row, slice));
        }
        #[cfg(target_arch = "wasm32")]
        {
            for (row, slice) in out_uninit.chunks_mut(cols).enumerate() {
                do_row(row, slice);
            }
        }
    } else {
        for (row, slice) in out_uninit.chunks_mut(cols).enumerate() {
            do_row(row, slice);
        }
    }

    Ok(combos)
}

#[cfg(feature = "python")]
#[pyfunction(name = "tilson")]
#[pyo3(signature = (data, period, volume_factor=None, kernel=None))]

pub fn tilson_py<'py>(
    py: Python<'py>,
    data: numpy::PyReadonlyArray1<'py, f64>,
    period: usize,
    volume_factor: Option<f64>,
    kernel: Option<&str>,
) -> PyResult<Bound<'py, numpy::PyArray1<f64>>> {
    use numpy::{IntoPyArray, PyArrayMethods};

    let slice_in = data.as_slice()?;
    let kern = validate_kernel(kernel, false)?;

    let params = TilsonParams {
        period: Some(period),
        volume_factor: volume_factor.or(Some(0.0)),
    };
    let tilson_in = TilsonInput::from_slice(slice_in, params);

    let result_vec: Vec<f64> = py
        .allow_threads(|| tilson_with_kernel(&tilson_in, kern).map(|o| o.values))
        .map_err(|e| PyValueError::new_err(e.to_string()))?;

    Ok(result_vec.into_pyarray(py))
}

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

#[cfg(feature = "python")]
#[pymethods]
impl TilsonStreamPy {
    #[new]
    fn new(period: usize, volume_factor: Option<f64>) -> PyResult<Self> {
        let params = TilsonParams {
            period: Some(period),
            volume_factor: volume_factor.or(Some(0.0)),
        };
        let stream =
            TilsonStream::try_new(params).map_err(|e| PyValueError::new_err(e.to_string()))?;
        Ok(TilsonStreamPy { stream })
    }

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

#[cfg(feature = "python")]
#[pyfunction(name = "tilson_batch")]
#[pyo3(signature = (data, period_range, volume_factor_range=None, kernel=None))]

pub fn tilson_batch_py<'py>(
    py: Python<'py>,
    data: numpy::PyReadonlyArray1<'py, f64>,
    period_range: (usize, usize, usize),
    volume_factor_range: Option<(f64, f64, f64)>,
    kernel: Option<&str>,
) -> PyResult<Bound<'py, pyo3::types::PyDict>> {
    use numpy::{IntoPyArray, PyArray1, PyArrayMethods};
    use pyo3::types::PyDict;

    let slice_in = data.as_slice()?;
    let kern = validate_kernel(kernel, true)?;

    let sweep = TilsonBatchRange {
        period: period_range,
        volume_factor: volume_factor_range.unwrap_or((0.0, 0.0, 0.0)),
    };

    let combos_dim = expand_grid(&sweep).map_err(|e| PyValueError::new_err(e.to_string()))?;
    let rows = combos_dim.len();
    let cols = slice_in.len();
    let total = rows
        .checked_mul(cols)
        .ok_or_else(|| PyValueError::new_err("dimensions too large to allocate"))?;

    let out_arr = unsafe { PyArray1::<f64>::new(py, [total], false) };
    let slice_out = unsafe { out_arr.as_slice_mut()? };

    let combos = py
        .allow_threads(|| {
            let kernel = match kern {
                Kernel::Auto => detect_best_batch_kernel(),
                k => k,
            };

            let simd = match kernel {
                Kernel::Avx512Batch => Kernel::Avx512,
                Kernel::Avx2Batch => Kernel::Avx2,
                Kernel::ScalarBatch => Kernel::Scalar,
                _ => unreachable!(),
            };

            tilson_batch_inner_into(slice_in, &sweep, simd, true, slice_out)
        })
        .map_err(|e| PyValueError::new_err(e.to_string()))?;

    let dict = PyDict::new(py);
    dict.set_item("values", out_arr.reshape((rows, cols))?)?;
    dict.set_item(
        "periods",
        combos
            .iter()
            .map(|p| p.period.unwrap() as u64)
            .collect::<Vec<_>>()
            .into_pyarray(py),
    )?;
    dict.set_item(
        "volume_factors",
        combos
            .iter()
            .map(|p| p.volume_factor.unwrap())
            .collect::<Vec<_>>()
            .into_pyarray(py),
    )?;
    Ok(dict)
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyfunction(name = "tilson_cuda_batch_dev")]
#[pyo3(signature = (data_f32, period_range, volume_factor_range=None, device_id=0))]
pub fn tilson_cuda_batch_dev_py(
    py: Python<'_>,
    data_f32: numpy::PyReadonlyArray1<'_, f32>,
    period_range: (usize, usize, usize),
    volume_factor_range: Option<(f64, f64, f64)>,
    device_id: usize,
) -> PyResult<DeviceArrayF32TilsonPy> {
    if !cuda_available() {
        return Err(PyValueError::new_err("CUDA not available"));
    }

    let slice_in = data_f32.as_slice()?;
    let sweep = TilsonBatchRange {
        period: period_range,
        volume_factor: volume_factor_range.unwrap_or((0.0, 0.0, 0.0)),
    };

    let inner = py.allow_threads(|| {
        let cuda = CudaTilson::new(device_id).map_err(|e| PyValueError::new_err(e.to_string()))?;
        cuda.tilson_batch_dev(slice_in, &sweep)
            .map_err(|e| PyValueError::new_err(e.to_string()))
    })?;

    Ok(DeviceArrayF32TilsonPy { inner })
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyfunction(name = "tilson_cuda_many_series_one_param_dev")]
#[pyo3(signature = (data_tm_f32, period, volume_factor, device_id=0))]
pub fn tilson_cuda_many_series_one_param_dev_py(
    py: Python<'_>,
    data_tm_f32: numpy::PyReadonlyArray2<'_, f32>,
    period: usize,
    volume_factor: f64,
    device_id: usize,
) -> PyResult<DeviceArrayF32TilsonPy> {
    if !cuda_available() {
        return Err(PyValueError::new_err("CUDA not available"));
    }

    use numpy::PyUntypedArrayMethods;

    let flat = data_tm_f32.as_slice()?;
    let rows = data_tm_f32.shape()[0];
    let cols = data_tm_f32.shape()[1];
    let params = TilsonParams {
        period: Some(period),
        volume_factor: Some(volume_factor),
    };

    let inner = py.allow_threads(|| {
        let cuda = CudaTilson::new(device_id).map_err(|e| PyValueError::new_err(e.to_string()))?;
        cuda.tilson_many_series_one_param_time_major_dev(flat, cols, rows, &params)
            .map_err(|e| PyValueError::new_err(e.to_string()))
    })?;

    Ok(DeviceArrayF32TilsonPy { inner })
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyclass(module = "ta_indicators.cuda", unsendable)]
pub struct DeviceArrayF32TilsonPy {
    pub(crate) inner: DeviceArrayF32Tilson,
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pymethods]
impl DeviceArrayF32TilsonPy {
    #[getter]
    fn __cuda_array_interface__<'py>(&self, py: Python<'py>) -> PyResult<Bound<'py, PyDict>> {
        let d = PyDict::new(py);
        d.set_item("shape", (self.inner.rows, self.inner.cols))?;
        d.set_item("typestr", "<f4")?;
        d.set_item(
            "strides",
            (
                self.inner.cols * std::mem::size_of::<f32>(),
                std::mem::size_of::<f32>(),
            ),
        )?;
        d.set_item("data", (self.inner.device_ptr() as usize, false))?;

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

    fn __dlpack_device__(&self) -> (i32, i32) {
        (2, self.inner.device_id as i32)
    }

    #[pyo3(signature = (stream=None, max_version=None, dl_device=None, copy=None))]
    fn __dlpack__<'py>(
        &mut self,
        py: Python<'py>,
        stream: Option<pyo3::PyObject>,
        max_version: Option<pyo3::PyObject>,
        dl_device: Option<pyo3::PyObject>,
        copy: Option<pyo3::PyObject>,
    ) -> PyResult<PyObject> {
        use crate::utilities::dlpack_cuda::export_f32_cuda_dlpack_2d;

        let (kdl, alloc_dev) = self.__dlpack_device__();
        if let Some(dev_obj) = dl_device.as_ref() {
            if let Ok((dev_ty, dev_id)) = dev_obj.extract::<(i32, i32)>(py) {
                if dev_ty != kdl || dev_id != alloc_dev {
                    let wants_copy = copy
                        .as_ref()
                        .and_then(|c| c.extract::<bool>(py).ok())
                        .unwrap_or(false);
                    if wants_copy {
                        return Err(PyValueError::new_err(
                            "device copy not implemented for __dlpack__",
                        ));
                    } else {
                        return Err(PyValueError::new_err("dl_device mismatch for __dlpack__"));
                    }
                }
            }
        }
        let _ = stream;

        let dummy =
            DeviceBuffer::from_slice(&[]).map_err(|e| PyValueError::new_err(e.to_string()))?;
        let ctx = self.inner.ctx.clone();
        let device_id = self.inner.device_id;
        let inner = std::mem::replace(
            &mut self.inner,
            DeviceArrayF32Tilson {
                buf: dummy,
                rows: 0,
                cols: 0,
                ctx,
                device_id,
            },
        );

        let rows = inner.rows;
        let cols = inner.cols;
        let buf = inner.buf;

        let max_version_bound = max_version.map(|obj| obj.into_bound(py));

        export_f32_cuda_dlpack_2d(py, buf, rows, cols, alloc_dev, max_version_bound)
    }
}

#[cfg(feature = "python")]
#[pyfunction(name = "tilson_into")]
#[pyo3(signature = (data, period, volume_factor=None, kernel=None))]
pub fn tilson_into_py<'py>(
    py: Python<'py>,
    data: numpy::PyReadonlyArray1<'py, f64>,
    period: usize,
    volume_factor: Option<f64>,
    kernel: Option<&str>,
) -> PyResult<Bound<'py, numpy::PyArray1<f64>>> {
    use numpy::{PyArray1, PyArrayMethods};
    let slice_in = data.as_slice()?;
    let out = unsafe { PyArray1::<f64>::new(py, [slice_in.len()], false) };
    let slice_out = unsafe { out.as_slice_mut()? };

    let kern = validate_kernel(kernel, false)?;
    let params = TilsonParams {
        period: Some(period),
        volume_factor: Some(volume_factor.unwrap_or(0.0)),
    };
    let input = TilsonInput::from_slice(slice_in, params);

    py.allow_threads(|| tilson_into_slice(slice_out, &input, kern))
        .map_err(|e| PyValueError::new_err(e.to_string()))?;
    Ok(out)
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[derive(Serialize, Deserialize)]
pub struct TilsonBatchConfig {
    pub period_range: (usize, usize, usize),
    pub volume_factor_range: (f64, f64, f64),
}

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

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen(js_name = tilson_js)]

pub fn tilson_js(data: &[f64], period: usize, volume_factor: f64) -> Result<Vec<f64>, JsValue> {
    let params = TilsonParams {
        period: Some(period),
        volume_factor: Some(volume_factor),
    };
    let input = TilsonInput::from_slice(data, params);
    let mut out = vec![0.0; data.len()];
    tilson_into_slice(&mut out, &input, Kernel::Auto)
        .map_err(|e| JsValue::from_str(&e.to_string()))?;
    Ok(out)
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen(js_name = tilson_batch_js)]

pub fn tilson_batch_js(
    data: &[f64],
    period_start: usize,
    period_end: usize,
    period_step: usize,
    v_factor_start: f64,
    v_factor_end: f64,
    v_factor_step: f64,
) -> Result<Vec<f64>, JsValue> {
    let sweep = TilsonBatchRange {
        period: (period_start, period_end, period_step),
        volume_factor: (v_factor_start, v_factor_end, v_factor_step),
    };

    let output = tilson_batch_with_kernel(data, &sweep, Kernel::ScalarBatch)
        .map_err(|e| JsValue::from_str(&e.to_string()))?;

    Ok(output.values)
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen(js_name = tilson_batch_metadata_js)]

pub fn tilson_batch_metadata_js(
    period_start: usize,
    period_end: usize,
    period_step: usize,
    v_factor_start: f64,
    v_factor_end: f64,
    v_factor_step: f64,
) -> Vec<f64> {
    let sweep = TilsonBatchRange {
        period: (period_start, period_end, period_step),
        volume_factor: (v_factor_start, v_factor_end, v_factor_step),
    };

    let combos = match expand_grid(&sweep) {
        Ok(v) => v,
        Err(_) => return Vec::new(),
    };
    let mut result = Vec::with_capacity(combos.len() * 2);

    for combo in &combos {
        result.push(combo.period.unwrap() as f64);
    }

    for combo in &combos {
        result.push(combo.volume_factor.unwrap());
    }

    result
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen(js_name = tilson_batch)]
pub fn tilson_batch_unified_js(data: &[f64], config: JsValue) -> Result<JsValue, JsValue> {
    let config: TilsonBatchConfig = serde_wasm_bindgen::from_value(config)
        .map_err(|e| JsValue::from_str(&format!("Invalid config: {}", e)))?;

    let sweep = TilsonBatchRange {
        period: config.period_range,
        volume_factor: config.volume_factor_range,
    };

    let output = tilson_batch_with_kernel(data, &sweep, Kernel::ScalarBatch)
        .map_err(|e| JsValue::from_str(&e.to_string()))?;

    let js_output = TilsonBatchJsOutput {
        values: output.values,
        combos: output.combos,
        rows: output.rows,
        cols: output.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 tilson_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 tilson_free(ptr: *mut f64, len: usize) {
    unsafe {
        let _ = Vec::from_raw_parts(ptr, len, len);
    }
}

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

    unsafe {
        let data = std::slice::from_raw_parts(in_ptr, len);

        if period == 0 || period > len {
            return Err(JsValue::from_str("Invalid period"));
        }

        let params = TilsonParams {
            period: Some(period),
            volume_factor: Some(volume_factor),
        };
        let input = TilsonInput::from_slice(data, params);

        let first = data
            .iter()
            .position(|&x| !x.is_nan())
            .ok_or_else(|| JsValue::from_str("All values are NaN"))?;

        let warmup = first + 6 * (period - 1);

        if in_ptr == out_ptr {
            let mut temp = vec![f64::NAN; len];
            tilson_compute_into(
                data,
                period,
                volume_factor,
                first,
                Kernel::Scalar,
                &mut temp,
            )
            .map_err(|e| JsValue::from_str(&e.to_string()))?;
            std::ptr::copy_nonoverlapping(temp.as_ptr(), out_ptr, len);
        } else {
            let out = std::slice::from_raw_parts_mut(out_ptr, len);

            for i in 0..warmup.min(len) {
                out[i] = f64::NAN;
            }
            tilson_compute_into(data, period, volume_factor, first, Kernel::Scalar, out)
                .map_err(|e| JsValue::from_str(&e.to_string()))?;
        }

        Ok(())
    }
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn tilson_batch_into(
    in_ptr: *const f64,
    out_ptr: *mut f64,
    len: usize,
    period_start: usize,
    period_end: usize,
    period_step: usize,
    v_factor_start: f64,
    v_factor_end: f64,
    v_factor_step: f64,
) -> Result<usize, JsValue> {
    if in_ptr.is_null() || out_ptr.is_null() {
        return Err(JsValue::from_str(
            "null pointer passed to tilson_batch_into",
        ));
    }

    unsafe {
        let data = std::slice::from_raw_parts(in_ptr, len);

        let sweep = TilsonBatchRange {
            period: (period_start, period_end, period_step),
            volume_factor: (v_factor_start, v_factor_end, v_factor_step),
        };

        let combos = expand_grid(&sweep).map_err(|e| JsValue::from_str(&e.to_string()))?;
        let rows = combos.len();
        let cols = len;
        let total = rows * cols;

        if total == 0 {
            return Err(JsValue::from_str("Invalid batch configuration"));
        }

        let out = std::slice::from_raw_parts_mut(out_ptr, total);

        tilson_batch_inner_into(data, &sweep, Kernel::Scalar, false, out)
            .map_err(|e| JsValue::from_str(&e.to_string()))?;

        Ok(rows)
    }
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
#[deprecated(
    since = "1.0.0",
    note = "For weight reuse patterns, use the fast/unsafe API with persistent buffers"
)]
pub struct TilsonContext {
    period: usize,
    c1: f64,
    c2: f64,
    c3: f64,
    c4: f64,
    kernel: Kernel,

    ema1: f64,
    ema2: f64,
    ema3: f64,
    ema4: f64,
    ema5: f64,
    ema6: f64,
    initialized: bool,
    warmup_count: usize,
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
#[allow(deprecated)]
impl TilsonContext {
    #[wasm_bindgen(constructor)]
    #[deprecated(
        since = "1.0.0",
        note = "For weight reuse patterns, use the fast/unsafe API with persistent buffers"
    )]
    pub fn new(period: usize, volume_factor: f64) -> Result<TilsonContext, JsValue> {
        if period == 0 {
            return Err(JsValue::from_str("Invalid period: 0"));
        }
        if volume_factor.is_nan() || volume_factor.is_infinite() {
            return Err(JsValue::from_str(&format!(
                "Invalid volume factor: {}",
                volume_factor
            )));
        }

        let c1 = -volume_factor.powi(3);
        let c2 = 3.0 * volume_factor.powi(2) + 3.0 * volume_factor.powi(3);
        let c3 = -6.0 * volume_factor.powi(2) - 3.0 * volume_factor - 3.0 * volume_factor.powi(3);
        let c4 = 1.0 + 3.0 * volume_factor + volume_factor.powi(3) + 3.0 * volume_factor.powi(2);

        Ok(TilsonContext {
            period,
            c1,
            c2,
            c3,
            c4,
            kernel: Kernel::Scalar,
            ema1: 0.0,
            ema2: 0.0,
            ema3: 0.0,
            ema4: 0.0,
            ema5: 0.0,
            ema6: 0.0,
            initialized: false,
            warmup_count: 0,
        })
    }

    #[wasm_bindgen]
    pub fn update(&mut self, value: f64) -> Option<f64> {
        if value.is_nan() {
            return None;
        }

        let alpha = 2.0 / (self.period as f64 + 1.0);

        if !self.initialized {
            self.ema1 = value;
            self.ema2 = value;
            self.ema3 = value;
            self.ema4 = value;
            self.ema5 = value;
            self.ema6 = value;
            self.initialized = true;
        } else {
            self.ema1 = alpha * value + (1.0 - alpha) * self.ema1;
            self.ema2 = alpha * self.ema1 + (1.0 - alpha) * self.ema2;
            self.ema3 = alpha * self.ema2 + (1.0 - alpha) * self.ema3;
            self.ema4 = alpha * self.ema3 + (1.0 - alpha) * self.ema4;
            self.ema5 = alpha * self.ema4 + (1.0 - alpha) * self.ema5;
            self.ema6 = alpha * self.ema5 + (1.0 - alpha) * self.ema6;
        }

        self.warmup_count += 1;

        if self.warmup_count <= 6 * (self.period - 1) {
            None
        } else {
            Some(
                self.c1 * self.ema6
                    + self.c2 * self.ema5
                    + self.c3 * self.ema4
                    + self.c4 * self.ema3,
            )
        }
    }

    #[wasm_bindgen]
    pub fn reset(&mut self) {
        self.ema1 = 0.0;
        self.ema2 = 0.0;
        self.ema3 = 0.0;
        self.ema4 = 0.0;
        self.ema5 = 0.0;
        self.ema6 = 0.0;
        self.initialized = false;
        self.warmup_count = 0;
    }

    #[wasm_bindgen]
    pub fn get_warmup_period(&self) -> usize {
        6 * (self.period - 1)
    }
}