vector_ta/indicators/

di.rs

#[cfg(all(feature = "python", feature = "cuda"))]
use numpy::PyUntypedArrayMethods;
#[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::*;

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 = "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 thiserror::Error;

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

#[derive(Debug, Clone)]
pub struct DiOutput {
    pub plus: Vec<f64>,
    pub minus: Vec<f64>,
}

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

impl Default for DiParams {
    fn default() -> Self {
        Self { period: Some(14) }
    }
}

#[derive(Debug, Clone)]
pub struct DiInput<'a> {
    pub data: DiData<'a>,
    pub params: DiParams,
}

impl<'a> DiInput<'a> {
    #[inline(always)]
    pub fn from_candles(candles: &'a Candles, params: DiParams) -> Self {
        Self {
            data: DiData::Candles { candles },
            params,
        }
    }
    #[inline(always)]
    pub fn from_slices(
        high: &'a [f64],
        low: &'a [f64],
        close: &'a [f64],
        params: DiParams,
    ) -> Self {
        Self {
            data: DiData::Slices { high, low, close },
            params,
        }
    }
    #[inline(always)]
    pub fn with_default_candles(candles: &'a Candles) -> Self {
        Self {
            data: DiData::Candles { candles },
            params: DiParams::default(),
        }
    }
    #[inline(always)]
    pub fn get_period(&self) -> usize {
        self.params.period.unwrap_or(14)
    }
    #[inline(always)]
    pub fn as_slices(&self) -> (&'a [f64], &'a [f64], &'a [f64]) {
        match &self.data {
            DiData::Candles { candles } => (
                source_type(candles, "high"),
                source_type(candles, "low"),
                source_type(candles, "close"),
            ),
            DiData::Slices { high, low, close } => (*high, *low, *close),
        }
    }
}

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

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

impl DiBuilder {
    #[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 kernel(mut self, k: Kernel) -> Self {
        self.kernel = k;
        self
    }
    #[inline(always)]
    pub fn apply(self, candles: &Candles) -> Result<DiOutput, DiError> {
        let params = DiParams {
            period: self.period,
        };
        let input = DiInput::from_candles(candles, params);
        di_with_kernel(&input, self.kernel)
    }
    #[inline(always)]
    pub fn apply_slices(
        self,
        high: &[f64],
        low: &[f64],
        close: &[f64],
    ) -> Result<DiOutput, DiError> {
        let params = DiParams {
            period: self.period,
        };
        let input = DiInput::from_slices(high, low, close, params);
        di_with_kernel(&input, self.kernel)
    }
    #[inline(always)]
    pub fn into_stream(self) -> Result<DiStream, DiError> {
        let params = DiParams {
            period: self.period,
        };
        DiStream::try_new(params)
    }
}

#[derive(Debug, Error)]
pub enum DiError {
    #[error("di: Empty data provided.")]
    EmptyInputData,
    #[error("di: Empty data provided for DI.")]
    EmptyData,
    #[error("di: Invalid period: period = {period}, data length = {data_len}")]
    InvalidPeriod { period: usize, data_len: usize },
    #[error("di: Not enough valid data: needed = {needed}, valid = {valid}")]
    NotEnoughValidData { needed: usize, valid: usize },
    #[error("di: All values are NaN.")]
    AllValuesNaN,
    #[error("di: Output length mismatch: expected {expected}, got {got}")]
    OutputLengthMismatch { expected: usize, got: usize },
    #[error("di: Invalid range expansion: start={start}, end={end}, step={step}")]
    InvalidRange {
        start: usize,
        end: usize,
        step: usize,
    },
    #[error("di: Invalid kernel for batch path: {0:?}")]
    InvalidKernelForBatch(Kernel),
}

#[inline]
pub fn di(input: &DiInput) -> Result<DiOutput, DiError> {
    di_with_kernel(input, Kernel::Auto)
}

#[inline(always)]
fn di_prepare<'a>(
    input: &'a DiInput<'a>,
    kernel: Kernel,
) -> Result<(&'a [f64], &'a [f64], &'a [f64], usize, usize, Kernel), DiError> {
    let (high, low, close) = match &input.data {
        DiData::Candles { candles } => {
            let h = source_type(candles, "high");
            let l = source_type(candles, "low");
            let c = source_type(candles, "close");
            (h, l, c)
        }
        DiData::Slices { high, low, close } => (*high, *low, *close),
    };
    let n = high.len();
    if n == 0 || low.len() != n || close.len() != n {
        return Err(DiError::EmptyInputData);
    }
    let period = input.get_period();
    if period == 0 || period > n {
        return Err(DiError::InvalidPeriod {
            period,
            data_len: n,
        });
    }
    let first_valid_idx =
        (0..n).find(|&i| !(high[i].is_nan() || low[i].is_nan() || close[i].is_nan()));
    let first_idx = match first_valid_idx {
        Some(idx) => idx,
        None => return Err(DiError::AllValuesNaN),
    };
    if (n - first_idx) < period {
        return Err(DiError::NotEnoughValidData {
            needed: period,
            valid: n - first_idx,
        });
    }

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

    Ok((high, low, close, period, first_idx, chosen))
}

pub fn di_with_kernel(input: &DiInput, kernel: Kernel) -> Result<DiOutput, DiError> {
    let (high, low, close, period, first_idx, chosen) = di_prepare(input, kernel)?;

    unsafe {
        match chosen {
            Kernel::Scalar | Kernel::ScalarBatch => di_scalar(high, low, close, period, first_idx),
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx2 | Kernel::Avx2Batch => di_avx2(high, low, close, period, first_idx),
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx512 | Kernel::Avx512Batch => di_avx512(high, low, close, period, first_idx),
            _ => unreachable!(),
        }
    }
}

#[cfg(not(all(feature = "nightly-avx", target_arch = "x86_64")))]
#[inline(always)]
pub unsafe fn di_avx2_into(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first_idx: usize,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) {
    di_scalar_into(high, low, close, period, first_idx, out_plus, out_minus)
}

#[cfg(not(all(feature = "nightly-avx", target_arch = "x86_64")))]
#[inline(always)]
pub unsafe fn di_avx512_into(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first_idx: usize,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) {
    di_scalar_into(high, low, close, period, first_idx, out_plus, out_minus)
}

#[inline(always)]
fn di_compute_into(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first_idx: usize,
    kernel: Kernel,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) {
    unsafe {
        match kernel {
            Kernel::Scalar | Kernel::ScalarBatch => {
                di_scalar_into(high, low, close, period, first_idx, out_plus, out_minus)
            }
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx2 | Kernel::Avx2Batch => {
                di_avx2_into(high, low, close, period, first_idx, out_plus, out_minus)
            }
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx512 | Kernel::Avx512Batch => {
                di_avx512_into(high, low, close, period, first_idx, out_plus, out_minus)
            }
            _ => unreachable!(),
        }
    }
}

pub fn di_into_slice(
    dst_plus: &mut [f64],
    dst_minus: &mut [f64],
    input: &DiInput,
    kern: Kernel,
) -> Result<(), DiError> {
    let (high, low, close, period, first_idx, chosen) = di_prepare(input, kern)?;

    let n = high.len();
    if dst_plus.len() != n || dst_minus.len() != n {
        return Err(DiError::OutputLengthMismatch {
            expected: n,
            got: dst_plus.len().min(dst_minus.len()),
        });
    }

    di_compute_into(
        high, low, close, period, first_idx, chosen, dst_plus, dst_minus,
    );

    let warmup_end = first_idx + period - 1;
    for v in &mut dst_plus[..warmup_end] {
        *v = f64::from_bits(0x7ff8_0000_0000_0000);
    }
    for v in &mut dst_minus[..warmup_end] {
        *v = f64::from_bits(0x7ff8_0000_0000_0000);
    }

    Ok(())
}

#[cfg(not(all(target_arch = "wasm32", feature = "wasm")))]
pub fn di_into(
    input: &DiInput,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) -> Result<(), DiError> {
    let (high, low, close, period, first_idx, chosen) = di_prepare(input, Kernel::Auto)?;

    let n = high.len();
    if out_plus.len() != n || out_minus.len() != n {
        return Err(DiError::OutputLengthMismatch {
            expected: n,
            got: out_plus.len().min(out_minus.len()),
        });
    }

    di_compute_into(
        high, low, close, period, first_idx, chosen, out_plus, out_minus,
    );

    let warmup_end = first_idx + period - 1;
    let qnan = f64::from_bits(0x7ff8_0000_0000_0000);
    for v in &mut out_plus[..warmup_end] {
        *v = qnan;
    }
    for v in &mut out_minus[..warmup_end] {
        *v = qnan;
    }

    Ok(())
}

#[inline(always)]
pub unsafe fn di_scalar_into(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first_idx: usize,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) {
    let n = high.len();
    if n == 0 {
        return;
    }

    let pf = period as f64;
    let invp = pf.recip();
    let keep = 1.0 - invp;

    let mut prev_h = high[first_idx];
    let mut prev_l = low[first_idx];
    let mut prev_c = close[first_idx];

    let start = first_idx + 1;
    let stop = first_idx + period;
    let mut plus_dm_sum = 0.0;
    let mut minus_dm_sum = 0.0;
    let mut tr_sum = 0.0;

    let mut i = start;
    while i < stop {
        let ch = high[i];
        let cl = low[i];
        let cc = close[i];

        let dp = ch - prev_h;
        let dm = prev_l - cl;
        if dp > dm && dp > 0.0 {
            plus_dm_sum += dp;
        }
        if dm > dp && dm > 0.0 {
            minus_dm_sum += dm;
        }

        let mut tr = ch - cl;
        let tr2 = (ch - prev_c).abs();
        let tr3 = (cl - prev_c).abs();
        if tr2 > tr {
            tr = tr2;
        }
        if tr3 > tr {
            tr = tr3;
        }
        tr_sum += tr;

        prev_h = ch;
        prev_l = cl;
        prev_c = cc;
        i += 1;
    }

    let mut cur_plus = plus_dm_sum;
    let mut cur_minus = minus_dm_sum;
    let mut cur_tr = tr_sum;

    let mut idx = stop - 1;
    let mut scale = if cur_tr == 0.0 { 0.0 } else { 100.0 / cur_tr };
    out_plus[idx] = cur_plus * scale;
    out_minus[idx] = cur_minus * scale;
    idx += 1;

    while idx < n {
        let ch = high[idx];
        let cl = low[idx];
        let cc = close[idx];

        let dp = ch - prev_h;
        let dm = prev_l - cl;
        let inc_p = if dp > dm && dp > 0.0 { dp } else { 0.0 };
        let inc_m = if dm > dp && dm > 0.0 { dm } else { 0.0 };

        cur_plus = cur_plus.mul_add(keep, inc_p);
        cur_minus = cur_minus.mul_add(keep, inc_m);

        let mut tr = ch - cl;
        let tr2 = (ch - prev_c).abs();
        let tr3 = (cl - prev_c).abs();
        if tr2 > tr {
            tr = tr2;
        }
        if tr3 > tr {
            tr = tr3;
        }
        cur_tr = cur_tr.mul_add(keep, tr);

        scale = if cur_tr == 0.0 { 0.0 } else { 100.0 / cur_tr };
        out_plus[idx] = cur_plus * scale;
        out_minus[idx] = cur_minus * scale;

        prev_h = ch;
        prev_l = cl;
        prev_c = cc;
        idx += 1;
    }
}

#[inline(always)]
pub unsafe fn di_scalar(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first_idx: usize,
) -> Result<DiOutput, DiError> {
    let n = high.len();
    let mut plus_di = alloc_with_nan_prefix(n, first_idx + period - 1);
    let mut minus_di = alloc_with_nan_prefix(n, first_idx + period - 1);
    di_scalar_into(
        high,
        low,
        close,
        period,
        first_idx,
        &mut plus_di,
        &mut minus_di,
    );
    Ok(DiOutput {
        plus: plus_di,
        minus: minus_di,
    })
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn di_avx2_into(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first_idx: usize,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) {
    di_scalar_into(high, low, close, period, first_idx, out_plus, out_minus)
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn di_avx2(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first_idx: usize,
) -> Result<DiOutput, DiError> {
    di_scalar(high, low, close, period, first_idx)
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn di_avx512_into(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first_idx: usize,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) {
    di_scalar_into(high, low, close, period, first_idx, out_plus, out_minus)
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn di_avx512(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first_idx: usize,
) -> Result<DiOutput, DiError> {
    di_scalar(high, low, close, period, first_idx)
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn di_avx512_short(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first_idx: usize,
) -> Result<DiOutput, DiError> {
    di_avx512(high, low, close, period, first_idx)
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn di_avx512_long(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first_idx: usize,
) -> Result<DiOutput, DiError> {
    di_avx512(high, low, close, period, first_idx)
}

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

    prev_h: f64,
    prev_l: f64,
    prev_c: f64,
    have_prev: bool,

    warm_plus: f64,
    warm_minus: f64,
    warm_tr: f64,
    warm_count: usize,

    cur_plus: f64,
    cur_minus: f64,
    cur_tr: f64,
    warmed: bool,
}

impl DiStream {
    pub fn try_new(params: DiParams) -> Result<Self, DiError> {
        let period = params.period.unwrap_or(14);
        if period == 0 {
            return Err(DiError::InvalidPeriod {
                period,
                data_len: 0,
            });
        }
        let pf = period as f64;
        Ok(Self {
            period,
            keep: 1.0 - pf.recip(),

            prev_h: f64::NAN,
            prev_l: f64::NAN,
            prev_c: f64::NAN,
            have_prev: false,

            warm_plus: 0.0,
            warm_minus: 0.0,
            warm_tr: 0.0,
            warm_count: 0,

            cur_plus: 0.0,
            cur_minus: 0.0,
            cur_tr: 0.0,
            warmed: false,
        })
    }

    #[inline(always)]
    fn reset(&mut self) {
        self.prev_h = f64::NAN;
        self.prev_l = f64::NAN;
        self.prev_c = f64::NAN;
        self.have_prev = false;

        self.warm_plus = 0.0;
        self.warm_minus = 0.0;
        self.warm_tr = 0.0;
        self.warm_count = 0;

        self.cur_plus = 0.0;
        self.cur_minus = 0.0;
        self.cur_tr = 0.0;
        self.warmed = false;
    }

    #[inline(always)]
    fn tr_fast(high: f64, low: f64, prev_close: f64) -> f64 {
        let hi = if high > prev_close { high } else { prev_close };
        let lo = if low < prev_close { low } else { prev_close };
        hi - lo
    }

    #[inline(always)]
    pub fn update(&mut self, high: f64, low: f64, close: f64) -> Option<(f64, f64)> {
        if high.is_nan() || low.is_nan() || close.is_nan() {
            self.reset();
            return None;
        }

        if !self.have_prev {
            self.prev_h = high;
            self.prev_l = low;
            self.prev_c = close;
            self.have_prev = true;
            self.warm_count = 1;
            return None;
        }

        let dp = high - self.prev_h;
        let dm = self.prev_l - low;

        let inc_p = if dp > dm && dp > 0.0 { dp } else { 0.0 };
        let inc_m = if dm > dp && dm > 0.0 { dm } else { 0.0 };
        let tr = Self::tr_fast(high, low, self.prev_c);

        self.prev_h = high;
        self.prev_l = low;
        self.prev_c = close;

        if !self.warmed {
            self.warm_plus += inc_p;
            self.warm_minus += inc_m;
            self.warm_tr += tr;
            self.warm_count += 1;

            if self.warm_count < self.period {
                return None;
            }

            self.cur_plus = self.warm_plus;
            self.cur_minus = self.warm_minus;
            self.cur_tr = self.warm_tr;
            self.warmed = true;

            let scale = if self.cur_tr == 0.0 {
                0.0
            } else {
                100.0 / self.cur_tr
            };
            return Some((self.cur_plus * scale, self.cur_minus * scale));
        }

        self.cur_plus = self.cur_plus.mul_add(self.keep, inc_p);
        self.cur_minus = self.cur_minus.mul_add(self.keep, inc_m);
        self.cur_tr = self.cur_tr.mul_add(self.keep, tr);

        let scale = if self.cur_tr == 0.0 {
            0.0
        } else {
            100.0 / self.cur_tr
        };
        Some((self.cur_plus * scale, self.cur_minus * scale))
    }
}

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

impl Default for DiBatchRange {
    fn default() -> Self {
        Self {
            period: (14, 263, 1),
        }
    }
}

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

impl DiBatchBuilder {
    pub fn new() -> Self {
        Self::default()
    }
    pub fn kernel(mut self, k: Kernel) -> Self {
        self.kernel = k;
        self
    }
    pub fn period_range(mut self, start: usize, end: usize, step: usize) -> Self {
        self.range.period = (start, end, step);
        self
    }
    pub fn period_static(mut self, p: usize) -> Self {
        self.range.period = (p, p, 0);
        self
    }
    pub fn apply_slices(
        self,
        high: &[f64],
        low: &[f64],
        close: &[f64],
    ) -> Result<DiBatchOutput, DiError> {
        di_batch_with_kernel(high, low, close, &self.range, self.kernel)
    }
    pub fn apply_candles(self, c: &Candles) -> Result<DiBatchOutput, DiError> {
        let h = source_type(c, "high");
        let l = source_type(c, "low");
        let cl = source_type(c, "close");
        self.apply_slices(h, l, cl)
    }
    pub fn with_default_candles(c: &Candles) -> Result<DiBatchOutput, DiError> {
        DiBatchBuilder::new().kernel(Kernel::Auto).apply_candles(c)
    }
}

pub fn di_batch_with_kernel(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    sweep: &DiBatchRange,
    k: Kernel,
) -> Result<DiBatchOutput, DiError> {
    let kernel = match k {
        Kernel::Auto => detect_best_batch_kernel(),
        other if other.is_batch() => other,
        other => return Err(DiError::InvalidKernelForBatch(other)),
    };
    let simd = match kernel {
        Kernel::Avx512Batch => Kernel::Avx512,
        Kernel::Avx2Batch => Kernel::Avx2,
        Kernel::ScalarBatch => Kernel::Scalar,
        _ => unreachable!(),
    };
    di_batch_par_slice(high, low, close, sweep, simd)
}

#[derive(Clone, Debug)]
pub struct DiBatchOutput {
    pub plus: Vec<f64>,
    pub minus: Vec<f64>,
    pub combos: Vec<DiParams>,
    pub rows: usize,
    pub cols: usize,
}
impl DiBatchOutput {
    pub fn row_for_params(&self, p: &DiParams) -> Option<usize> {
        self.combos
            .iter()
            .position(|c| c.period.unwrap_or(14) == p.period.unwrap_or(14))
    }
    pub fn plus_for(&self, p: &DiParams) -> Option<&[f64]> {
        self.row_for_params(p).map(|row| {
            let start = row * self.cols;
            &self.plus[start..start + self.cols]
        })
    }
    pub fn minus_for(&self, p: &DiParams) -> Option<&[f64]> {
        self.row_for_params(p).map(|row| {
            let start = row * self.cols;
            &self.minus[start..start + self.cols]
        })
    }
}

#[inline(always)]
fn expand_grid(r: &DiBatchRange) -> Vec<DiParams> {
    fn axis_usize((start, end, step): (usize, usize, usize)) -> Vec<usize> {
        if step == 0 || start == end {
            return vec![start];
        }
        if start < end {
            return (start..=end).step_by(step.max(1)).collect();
        }

        let mut v = Vec::new();
        let mut cur = start;
        while cur >= end {
            v.push(cur);
            if cur == end {
                break;
            }
            cur = cur.saturating_sub(step.max(1));
            if cur < end {
                break;
            }
        }
        v
    }
    let periods = axis_usize(r.period);
    let mut out = Vec::with_capacity(periods.len());
    for &p in &periods {
        out.push(DiParams { period: Some(p) });
    }
    out
}

#[inline(always)]
pub fn di_batch_slice(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    sweep: &DiBatchRange,
    kern: Kernel,
) -> Result<DiBatchOutput, DiError> {
    di_batch_inner(high, low, close, sweep, kern, false)
}

#[inline(always)]
pub fn di_batch_par_slice(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    sweep: &DiBatchRange,
    kern: Kernel,
) -> Result<DiBatchOutput, DiError> {
    di_batch_inner(high, low, close, sweep, kern, true)
}

#[inline(always)]
fn di_batch_inner_into(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    sweep: &DiBatchRange,
    kern: Kernel,
    parallel: bool,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) -> Result<Vec<DiParams>, DiError> {
    let combos = expand_grid(sweep);
    if combos.is_empty() {
        let (s, e, st) = sweep.period;
        return Err(DiError::InvalidRange {
            start: s,
            end: e,
            step: st,
        });
    }
    let n = high.len();
    if n == 0 || low.len() != n || close.len() != n {
        return Err(DiError::EmptyInputData);
    }
    let first = (0..n)
        .find(|&i| !(high[i].is_nan() || low[i].is_nan() || close[i].is_nan()))
        .ok_or(DiError::AllValuesNaN)?;
    let max_p = combos.iter().map(|c| c.period.unwrap()).max().unwrap();
    if n - first < max_p {
        return Err(DiError::NotEnoughValidData {
            needed: max_p,
            valid: n - first,
        });
    }

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

    unsafe {
        let total = rows.checked_mul(cols).ok_or(DiError::InvalidRange {
            start: sweep.period.0,
            end: sweep.period.1,
            step: sweep.period.2,
        })?;
        let plus_mu = std::slice::from_raw_parts_mut(
            out_plus.as_mut_ptr() as *mut std::mem::MaybeUninit<f64>,
            total,
        );
        let minus_mu = std::slice::from_raw_parts_mut(
            out_minus.as_mut_ptr() as *mut std::mem::MaybeUninit<f64>,
            total,
        );
        let warm: Vec<usize> = combos
            .iter()
            .map(|c| first.saturating_add(c.period.unwrap()).saturating_sub(1))
            .collect();
        init_matrix_prefixes(plus_mu, cols, &warm);
        init_matrix_prefixes(minus_mu, cols, &warm);
    }

    let mut up = vec![0.0f64; n];
    let mut dn = vec![0.0f64; n];
    let mut tr = vec![0.0f64; n];
    {
        let mut prev_h = high[first];
        let mut prev_l = low[first];
        let mut prev_c = close[first];
        let mut i = first + 1;
        while i < n {
            let ch = high[i];
            let cl = low[i];
            let dp = ch - prev_h;
            let dm = prev_l - cl;
            if dp > dm && dp > 0.0 {
                up[i] = dp;
            }
            if dm > dp && dm > 0.0 {
                dn[i] = dm;
            }
            let mut t = ch - cl;
            let t2 = (ch - prev_c).abs();
            let t3 = (cl - prev_c).abs();
            if t2 > t {
                t = t2;
            }
            if t3 > t {
                t = t3;
            }
            tr[i] = t;
            prev_h = ch;
            prev_l = cl;
            prev_c = close[i];
            i += 1;
        }
    }

    let do_row = |row: usize, out_plus: &mut [f64], out_minus: &mut [f64]| unsafe {
        let period = combos[row].period.unwrap();
        let result = di_row_scalar_precomputed(&up, &dn, &tr, period, first, out_plus, out_minus);
        debug_assert!(result.is_ok());
    };

    if parallel {
        #[cfg(not(target_arch = "wasm32"))]
        {
            out_plus
                .par_chunks_mut(cols)
                .zip(out_minus.par_chunks_mut(cols))
                .enumerate()
                .for_each(|(row, (pl, mi))| do_row(row, pl, mi));
        }

        #[cfg(target_arch = "wasm32")]
        {
            for (row, (pl, mi)) in out_plus
                .chunks_mut(cols)
                .zip(out_minus.chunks_mut(cols))
                .enumerate()
            {
                do_row(row, pl, mi);
            }
        }
    } else {
        for (row, (pl, mi)) in out_plus
            .chunks_mut(cols)
            .zip(out_minus.chunks_mut(cols))
            .enumerate()
        {
            do_row(row, pl, mi);
        }
    }

    Ok(combos)
}

#[inline(always)]
fn di_batch_inner(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    sweep: &DiBatchRange,
    kern: Kernel,
    parallel: bool,
) -> Result<DiBatchOutput, DiError> {
    let combos = expand_grid(sweep);
    if combos.is_empty() {
        let (s, e, st) = sweep.period;
        return Err(DiError::InvalidRange {
            start: s,
            end: e,
            step: st,
        });
    }
    let n = high.len();
    if n == 0 || low.len() != n || close.len() != n {
        return Err(DiError::EmptyInputData);
    }
    let first = (0..n)
        .find(|&i| !(high[i].is_nan() || low[i].is_nan() || close[i].is_nan()))
        .ok_or(DiError::AllValuesNaN)?;
    let max_p = combos.iter().map(|c| c.period.unwrap()).max().unwrap();
    if n - first < max_p {
        return Err(DiError::NotEnoughValidData {
            needed: max_p,
            valid: n - first,
        });
    }

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

    let _ = rows.checked_mul(cols).ok_or(DiError::InvalidRange {
        start: sweep.period.0,
        end: sweep.period.1,
        step: sweep.period.2,
    })?;

    let warmup_periods: Vec<usize> = combos
        .iter()
        .map(|c| first.saturating_add(c.period.unwrap()).saturating_sub(1))
        .collect();

    let mut plus_mu = make_uninit_matrix(rows, cols);
    let mut minus_mu = make_uninit_matrix(rows, cols);

    init_matrix_prefixes(&mut plus_mu, cols, &warmup_periods);
    init_matrix_prefixes(&mut minus_mu, cols, &warmup_periods);

    let mut plus_guard = core::mem::ManuallyDrop::new(plus_mu);
    let mut minus_guard = core::mem::ManuallyDrop::new(minus_mu);
    let plus: &mut [f64] = unsafe {
        core::slice::from_raw_parts_mut(plus_guard.as_mut_ptr() as *mut f64, plus_guard.len())
    };
    let minus: &mut [f64] = unsafe {
        core::slice::from_raw_parts_mut(minus_guard.as_mut_ptr() as *mut f64, minus_guard.len())
    };

    let mut up = vec![0.0f64; n];
    let mut dn = vec![0.0f64; n];
    let mut tr = vec![0.0f64; n];
    {
        let mut prev_h = high[first];
        let mut prev_l = low[first];
        let mut prev_c = close[first];
        let mut i = first + 1;
        while i < n {
            let ch = high[i];
            let cl = low[i];
            let dp = ch - prev_h;
            let dm = prev_l - cl;
            if dp > dm && dp > 0.0 {
                up[i] = dp;
            }
            if dm > dp && dm > 0.0 {
                dn[i] = dm;
            }
            let mut t = ch - cl;
            let t2 = (ch - prev_c).abs();
            let t3 = (cl - prev_c).abs();
            if t2 > t {
                t = t2;
            }
            if t3 > t {
                t = t3;
            }
            tr[i] = t;
            prev_h = ch;
            prev_l = cl;
            prev_c = close[i];
            i += 1;
        }
    }

    let do_row = |row: usize, out_plus: &mut [f64], out_minus: &mut [f64]| unsafe {
        let period = combos[row].period.unwrap();
        let result = di_row_scalar_precomputed(&up, &dn, &tr, period, first, out_plus, out_minus);
        debug_assert!(result.is_ok());
    };

    if parallel {
        #[cfg(not(target_arch = "wasm32"))]
        {
            plus.par_chunks_mut(cols)
                .zip(minus.par_chunks_mut(cols))
                .enumerate()
                .for_each(|(row, (pl, mi))| do_row(row, pl, mi));
        }

        #[cfg(target_arch = "wasm32")]
        {
            for (row, (pl, mi)) in plus
                .chunks_mut(cols)
                .zip(minus.chunks_mut(cols))
                .enumerate()
            {
                do_row(row, pl, mi);
            }
        }
    } else {
        for (row, (pl, mi)) in plus
            .chunks_mut(cols)
            .zip(minus.chunks_mut(cols))
            .enumerate()
        {
            do_row(row, pl, mi);
        }
    }

    let plus = unsafe {
        Vec::from_raw_parts(
            plus_guard.as_mut_ptr() as *mut f64,
            plus_guard.len(),
            plus_guard.capacity(),
        )
    };
    let minus = unsafe {
        Vec::from_raw_parts(
            minus_guard.as_mut_ptr() as *mut f64,
            minus_guard.len(),
            minus_guard.capacity(),
        )
    };

    Ok(DiBatchOutput {
        plus,
        minus,
        combos,
        rows,
        cols,
    })
}

#[inline(always)]
pub unsafe fn di_row_scalar(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first: usize,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) -> Result<(), DiError> {
    let n = high.len();
    if n == 0 {
        return Ok(());
    }

    let pf = period as f64;
    let invp = pf.recip();
    let keep = 1.0 - invp;

    let mut prev_h = high[first];
    let mut prev_l = low[first];
    let mut prev_c = close[first];

    let start = first + 1;
    let stop = first + period;
    let mut plus_dm_sum = 0.0;
    let mut minus_dm_sum = 0.0;
    let mut tr_sum = 0.0;

    let mut i = start;
    while i < stop {
        let ch = high[i];
        let cl = low[i];
        let cc = close[i];

        let dp = ch - prev_h;
        let dm = prev_l - cl;
        if dp > dm && dp > 0.0 {
            plus_dm_sum += dp;
        }
        if dm > dp && dm > 0.0 {
            minus_dm_sum += dm;
        }

        let mut tr = ch - cl;
        let tr2 = (ch - prev_c).abs();
        let tr3 = (cl - prev_c).abs();
        if tr2 > tr {
            tr = tr2;
        }
        if tr3 > tr {
            tr = tr3;
        }
        tr_sum += tr;

        prev_h = ch;
        prev_l = cl;
        prev_c = cc;
        i += 1;
    }

    let mut cur_plus = plus_dm_sum;
    let mut cur_minus = minus_dm_sum;
    let mut cur_tr = tr_sum;

    let mut idx = stop - 1;
    let mut scale = if cur_tr == 0.0 { 0.0 } else { 100.0 / cur_tr };
    out_plus[idx] = cur_plus * scale;
    out_minus[idx] = cur_minus * scale;
    idx += 1;

    while idx < n {
        let ch = high[idx];
        let cl = low[idx];
        let cc = close[idx];

        let dp = ch - prev_h;
        let dm = prev_l - cl;
        let inc_p = if dp > dm && dp > 0.0 { dp } else { 0.0 };
        let inc_m = if dm > dp && dm > 0.0 { dm } else { 0.0 };

        cur_plus = cur_plus.mul_add(keep, inc_p);
        cur_minus = cur_minus.mul_add(keep, inc_m);

        let mut tr = ch - cl;
        let tr2 = (ch - prev_c).abs();
        let tr3 = (cl - prev_c).abs();
        if tr2 > tr {
            tr = tr2;
        }
        if tr3 > tr {
            tr = tr3;
        }
        cur_tr = cur_tr.mul_add(keep, tr);

        scale = if cur_tr == 0.0 { 0.0 } else { 100.0 / cur_tr };
        out_plus[idx] = cur_plus * scale;
        out_minus[idx] = cur_minus * scale;

        prev_h = ch;
        prev_l = cl;
        prev_c = cc;
        idx += 1;
    }
    Ok(())
}

#[inline(always)]
pub unsafe fn di_row_scalar_precomputed(
    up: &[f64],
    dn: &[f64],
    tr: &[f64],
    period: usize,
    first: usize,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) -> Result<(), DiError> {
    let n = up.len();
    if n == 0 {
        return Ok(());
    }

    let pf = period as f64;
    let invp = pf.recip();
    let keep = 1.0 - invp;

    let start = first + 1;
    let stop = first + period;
    let mut plus_dm_sum = 0.0;
    let mut minus_dm_sum = 0.0;
    let mut tr_sum = 0.0;

    let mut i = start;
    while i < stop {
        plus_dm_sum += up[i];
        minus_dm_sum += dn[i];
        tr_sum += tr[i];
        i += 1;
    }

    let mut cur_plus = plus_dm_sum;
    let mut cur_minus = minus_dm_sum;
    let mut cur_tr = tr_sum;

    let mut idx = stop - 1;
    let mut scale = if cur_tr == 0.0 { 0.0 } else { 100.0 / cur_tr };
    out_plus[idx] = cur_plus * scale;
    out_minus[idx] = cur_minus * scale;
    idx += 1;

    while idx < n {
        cur_plus = cur_plus.mul_add(keep, up[idx]);
        cur_minus = cur_minus.mul_add(keep, dn[idx]);
        cur_tr = cur_tr.mul_add(keep, tr[idx]);
        scale = if cur_tr == 0.0 { 0.0 } else { 100.0 / cur_tr };
        out_plus[idx] = cur_plus * scale;
        out_minus[idx] = cur_minus * scale;
        idx += 1;
    }
    Ok(())
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn di_row_avx2(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first: usize,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) -> Result<(), DiError> {
    di_row_scalar(high, low, close, period, first, out_plus, out_minus)
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn di_row_avx512(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first: usize,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) -> Result<(), DiError> {
    di_row_scalar(high, low, close, period, first, out_plus, out_minus)
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn di_row_avx512_short(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first: usize,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) -> Result<(), DiError> {
    di_row_avx512(high, low, close, period, first, out_plus, out_minus)
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn di_row_avx512_long(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    period: usize,
    first: usize,
    out_plus: &mut [f64],
    out_minus: &mut [f64],
) -> Result<(), DiError> {
    di_row_avx512(high, low, close, period, first, out_plus, out_minus)
}

#[inline(always)]
fn true_range(current_high: f64, current_low: f64, prev_close: f64) -> f64 {
    let mut tr1 = current_high - current_low;
    let tr2 = (current_high - prev_close).abs();
    let tr3 = (current_low - prev_close).abs();
    if tr2 > tr1 {
        tr1 = tr2;
    }
    if tr3 > tr1 {
        tr1 = tr3;
    }
    tr1
}

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

    fn check_di_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 = DiParams { period: None };
        let input_default = DiInput::from_candles(&candles, default_params);
        let output_default = di_with_kernel(&input_default, kernel)?;
        assert_eq!(output_default.plus.len(), candles.close.len());
        assert_eq!(output_default.minus.len(), candles.close.len());
        let params_period_10 = DiParams { period: Some(10) };
        let input_period_10 = DiInput::from_candles(&candles, params_period_10);
        let output_period_10 = di_with_kernel(&input_period_10, kernel)?;
        assert_eq!(output_period_10.plus.len(), candles.close.len());
        assert_eq!(output_period_10.minus.len(), candles.close.len());
        Ok(())
    }
    fn check_di_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 params = DiParams { period: Some(14) };
        let input = DiInput::from_candles(&candles, params);
        let di_result = di_with_kernel(&input, kernel)?;
        assert_eq!(di_result.plus.len(), candles.close.len());
        assert_eq!(di_result.minus.len(), candles.close.len());
        let test_plus = [
            10.99067007335658,
            11.306993269828585,
            10.948661818939213,
            10.683207768215592,
            9.802180952619183,
        ];
        let test_minus = [
            28.06728094177839,
            27.331240567633152,
            27.759989125359493,
            26.951434842917386,
            30.748897303623057,
        ];
        if di_result.plus.len() > 5 {
            let plus_tail = &di_result.plus[di_result.plus.len() - 5..];
            let minus_tail = &di_result.minus[di_result.minus.len() - 5..];
            for i in 0..5 {
                assert!(
                    (plus_tail[i] - test_plus[i]).abs() < 1e-6,
                    "Mismatch in +DI at tail index {}: expected {}, got {}",
                    i,
                    test_plus[i],
                    plus_tail[i]
                );
                assert!(
                    (minus_tail[i] - test_minus[i]).abs() < 1e-6,
                    "Mismatch in -DI at tail index {}: expected {}, got {}",
                    i,
                    test_minus[i],
                    minus_tail[i]
                );
            }
        }
        Ok(())
    }
    fn check_di_with_zero_period(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let high = [10.0, 11.0, 12.0];
        let low = [9.0, 8.0, 7.0];
        let close = [9.5, 10.0, 11.0];
        let params = DiParams { period: Some(0) };
        let input = DiInput::from_slices(&high, &low, &close, params);
        let result = di_with_kernel(&input, kernel);
        assert!(result.is_err());
        Ok(())
    }
    fn check_di_with_period_exceeding_data_length(
        test_name: &str,
        kernel: Kernel,
    ) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let high = [10.0, 11.0, 12.0];
        let low = [9.0, 8.0, 7.0];
        let close = [9.5, 10.0, 11.0];
        let params = DiParams { period: Some(10) };
        let input = DiInput::from_slices(&high, &low, &close, params);
        let result = di_with_kernel(&input, kernel);
        assert!(result.is_err());
        Ok(())
    }
    fn check_di_very_small_data_set(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let high = [42.0];
        let low = [41.0];
        let close = [41.5];
        let params = DiParams { period: Some(14) };
        let input = DiInput::from_slices(&high, &low, &close, params);
        let result = di_with_kernel(&input, kernel);
        assert!(result.is_err());
        Ok(())
    }
    fn check_di_with_slice_data_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 = DiParams { period: Some(14) };
        let first_input = DiInput::from_candles(&candles, first_params);
        let first_result = di_with_kernel(&first_input, kernel)?;
        assert_eq!(first_result.plus.len(), candles.close.len());
        assert_eq!(first_result.minus.len(), candles.close.len());
        let second_params = DiParams { period: Some(14) };
        let second_input = DiInput::from_slices(
            &first_result.plus,
            &first_result.minus,
            &candles.close,
            second_params,
        );
        let second_result = di_with_kernel(&second_input, kernel)?;
        assert_eq!(second_result.plus.len(), first_result.plus.len());
        assert_eq!(second_result.minus.len(), first_result.minus.len());
        Ok(())
    }
    fn check_di_accuracy_nan_check(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 params = DiParams { period: Some(14) };
        let input = DiInput::from_candles(&candles, params);
        let di_result = di_with_kernel(&input, kernel)?;
        assert_eq!(di_result.plus.len(), candles.close.len());
        assert_eq!(di_result.minus.len(), candles.close.len());
        if di_result.plus.len() > 40 {
            for i in 40..di_result.plus.len() {
                assert!(!di_result.plus[i].is_nan());
                assert!(!di_result.minus[i].is_nan());
            }
        }
        Ok(())
    }
    #[cfg(debug_assertions)]
    fn check_di_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_params = vec![
            DiParams::default(),
            DiParams { period: Some(2) },
            DiParams { period: Some(5) },
            DiParams { period: Some(7) },
            DiParams { period: Some(10) },
            DiParams { period: Some(20) },
            DiParams { period: Some(30) },
            DiParams { period: Some(50) },
            DiParams { period: Some(100) },
            DiParams { period: Some(200) },
        ];

        for (param_idx, params) in test_params.iter().enumerate() {
            let input = DiInput::from_candles(&candles, params.clone());
            let output = di_with_kernel(&input, kernel)?;

            for (i, &val) in output.plus.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 plus output with params: {:?} (param set {})",
                        test_name, val, bits, i, params, param_idx
                    );
                }

                if bits == 0x22222222_22222222 {
                    panic!(
                        "[{}] Found init_matrix_prefixes poison value {} (0x{:016X}) at index {} \
						 in plus output with params: {:?} (param set {})",
                        test_name, val, bits, i, params, param_idx
                    );
                }

                if bits == 0x33333333_33333333 {
                    panic!(
                        "[{}] Found make_uninit_matrix poison value {} (0x{:016X}) at index {} \
						 in plus output with params: {:?} (param set {})",
                        test_name, val, bits, i, params, param_idx
                    );
                }
            }

            for (i, &val) in output.minus.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 minus output with params: {:?} (param set {})",
                        test_name, val, bits, i, params, param_idx
                    );
                }

                if bits == 0x22222222_22222222 {
                    panic!(
                        "[{}] Found init_matrix_prefixes poison value {} (0x{:016X}) at index {} \
						 in minus output with params: {:?} (param set {})",
                        test_name, val, bits, i, params, param_idx
                    );
                }

                if bits == 0x33333333_33333333 {
                    panic!(
                        "[{}] Found make_uninit_matrix poison value {} (0x{:016X}) at index {} \
						 in minus output with params: {:?} (param set {})",
                        test_name, val, bits, i, params, param_idx
                    );
                }
            }
        }

        Ok(())
    }

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

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

        let strat = (2usize..=50)
            .prop_flat_map(|period| {
                (
                    100.0f64..5000.0f64,
                    (period + 20)..400,
                    0.001f64..0.05f64,
                    -0.01f64..0.01f64,
                    Just(period),
                )
            })
            .prop_map(|(base_price, data_len, volatility, trend, period)| {
                let mut high = Vec::with_capacity(data_len);
                let mut low = Vec::with_capacity(data_len);
                let mut close = Vec::with_capacity(data_len);

                let mut price = base_price;

                for i in 0..data_len {
                    let trend_component = trend * i as f64;
                    let random_component = ((i * 137 + 11) % 100) as f64 / 100.0 - 0.5;
                    price = price * (1.0 + trend_component + random_component * volatility);

                    price = price.max(1.0);

                    let daily_range = price * volatility * (1.0 + ((i * 73) % 50) as f64 / 100.0);
                    let h = price + daily_range * 0.5;
                    let l = price - daily_range * 0.5;

                    let close_factor = ((i * 29 + 7) % 100) as f64 / 100.0;
                    let c = l + (h - l) * close_factor;

                    high.push(h);
                    low.push(l);
                    close.push(c);
                }

                (high, low, close, period)
            });

        proptest::test_runner::TestRunner::default()
            .run(&strat, |(high, low, close, period)| {
                let params = DiParams {
                    period: Some(period),
                };
                let input = DiInput::from_slices(&high, &low, &close, params);

                let output = di_with_kernel(&input, kernel)?;

                let ref_output = di_with_kernel(&input, Kernel::Scalar)?;

                prop_assert_eq!(output.plus.len(), high.len());
                prop_assert_eq!(output.minus.len(), high.len());

                let warmup_end = period - 1;
                for i in 0..warmup_end {
                    prop_assert!(
                        output.plus[i].is_nan(),
                        "Expected NaN at index {} during warmup, got {}",
                        i,
                        output.plus[i]
                    );
                    prop_assert!(
                        output.minus[i].is_nan(),
                        "Expected NaN at index {} during warmup, got {}",
                        i,
                        output.minus[i]
                    );
                }

                for i in warmup_end..high.len() {
                    let plus_val = output.plus[i];
                    let minus_val = output.minus[i];

                    prop_assert!(
                        plus_val.is_finite(),
                        "Expected finite +DI at index {}, got {}",
                        i,
                        plus_val
                    );
                    prop_assert!(
                        minus_val.is_finite(),
                        "Expected finite -DI at index {}, got {}",
                        i,
                        minus_val
                    );

                    prop_assert!(
                        plus_val >= 0.0 && plus_val <= 100.0,
                        "+DI at index {} = {} is out of range [0, 100]",
                        i,
                        plus_val
                    );
                    prop_assert!(
                        minus_val >= 0.0 && minus_val <= 100.0,
                        "-DI at index {} = {} is out of range [0, 100]",
                        i,
                        minus_val
                    );
                }

                for i in 0..high.len() {
                    let plus_val = output.plus[i];
                    let minus_val = output.minus[i];
                    let ref_plus = ref_output.plus[i];
                    let ref_minus = ref_output.minus[i];

                    if plus_val.is_nan() || ref_plus.is_nan() {
                        prop_assert_eq!(
                            plus_val.is_nan(),
                            ref_plus.is_nan(),
                            "NaN mismatch in +DI at index {}",
                            i
                        );
                    } else {
                        prop_assert!(
                            (plus_val - ref_plus).abs() <= 1e-9,
                            "+DI mismatch at index {}: {} vs {} (diff: {})",
                            i,
                            plus_val,
                            ref_plus,
                            (plus_val - ref_plus).abs()
                        );
                    }

                    if minus_val.is_nan() || ref_minus.is_nan() {
                        prop_assert_eq!(
                            minus_val.is_nan(),
                            ref_minus.is_nan(),
                            "NaN mismatch in -DI at index {}",
                            i
                        );
                    } else {
                        prop_assert!(
                            (minus_val - ref_minus).abs() <= 1e-9,
                            "-DI mismatch at index {}: {} vs {} (diff: {})",
                            i,
                            minus_val,
                            ref_minus,
                            (minus_val - ref_minus).abs()
                        );
                    }
                }

                let constant_high = vec![100.0; 50];
                let constant_low = vec![100.0; 50];
                let constant_close = vec![100.0; 50];
                let const_params = DiParams {
                    period: Some(period),
                };
                let const_input = DiInput::from_slices(
                    &constant_high,
                    &constant_low,
                    &constant_close,
                    const_params,
                );

                if let Ok(const_output) = di_with_kernel(&const_input, kernel) {
                    for i in (period + 5)..constant_high.len() {
                        prop_assert!(
                            const_output.plus[i] < 1.0,
                            "Expected near-zero +DI for constant prices, got {} at index {}",
                            const_output.plus[i],
                            i
                        );
                        prop_assert!(
                            const_output.minus[i] < 1.0,
                            "Expected near-zero -DI for constant prices, got {} at index {}",
                            const_output.minus[i],
                            i
                        );
                    }
                }

                let mut spike_high = vec![100.0; 30];
                let mut spike_low = vec![99.0; 30];
                let mut spike_close = vec![99.5; 30];

                spike_high[15] = 120.0;
                spike_low[15] = 80.0;
                spike_close[15] = 100.0;

                let spike_params = DiParams {
                    period: Some(period.min(10)),
                };
                let spike_input =
                    DiInput::from_slices(&spike_high, &spike_low, &spike_close, spike_params);

                if let Ok(spike_output) = di_with_kernel(&spike_input, kernel) {
                    for (i, (&plus, &minus)) in spike_output
                        .plus
                        .iter()
                        .zip(spike_output.minus.iter())
                        .enumerate()
                    {
                        if !plus.is_nan() {
                            prop_assert!(
                                plus >= 0.0 && plus <= 100.0,
                                "Volatility spike caused +DI out of range at {}: {}",
                                i,
                                plus
                            );
                        }
                        if !minus.is_nan() {
                            prop_assert!(
                                minus >= 0.0 && minus <= 100.0,
                                "Volatility spike caused -DI out of range at {}: {}",
                                i,
                                minus
                            );
                        }
                    }
                }

                if period <= 20 {
                    let trend_len = 200;
                    let mut uptrend_high = Vec::with_capacity(trend_len);
                    let mut uptrend_low = Vec::with_capacity(trend_len);
                    let mut uptrend_close = Vec::with_capacity(trend_len);

                    for i in 0..trend_len {
                        let price = 100.0 + i as f64 * 2.0;
                        uptrend_high.push(price + 1.0);
                        uptrend_low.push(price - 1.0);
                        uptrend_close.push(price);
                    }

                    let trend_params = DiParams {
                        period: Some(period),
                    };
                    let trend_input = DiInput::from_slices(
                        &uptrend_high,
                        &uptrend_low,
                        &uptrend_close,
                        trend_params,
                    );

                    if let Ok(trend_output) = di_with_kernel(&trend_input, kernel) {
                        let check_start = trend_len / 2;
                        let mut plus_wins = 0;
                        let mut minus_wins = 0;

                        for i in check_start..trend_len {
                            if trend_output.plus[i] > trend_output.minus[i] {
                                plus_wins += 1;
                            } else {
                                minus_wins += 1;
                            }
                        }

                        if plus_wins + minus_wins > 0 {
                            prop_assert!(
								plus_wins > minus_wins,
								"In uptrend, expected +DI > -DI more often. +DI wins: {}, -DI wins: {} (period: {})",
								plus_wins, minus_wins, period
							);
                        }
                    }
                }

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

        Ok(())
    }

    #[test]
    fn test_di_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 params = DiParams::default();
        let input = DiInput::from_candles(&candles, params);

        let baseline = di(&input)?;

        let n = candles.close.len();
        let mut plus = vec![0.0_f64; n];
        let mut minus = vec![0.0_f64; n];
        di_into(&input, &mut plus, &mut minus)?;

        assert_eq!(baseline.plus.len(), plus.len());
        assert_eq!(baseline.minus.len(), minus.len());

        fn eq_or_both_nan(a: f64, b: f64) -> bool {
            (a.is_nan() && b.is_nan()) || (a == b) || ((a - b).abs() <= 1e-12)
        }

        for i in 0..n {
            assert!(
                eq_or_both_nan(baseline.plus[i], plus[i]),
                "+DI mismatch at index {}: baseline={} vs into={}",
                i,
                baseline.plus[i],
                plus[i]
            );
            assert!(
                eq_or_both_nan(baseline.minus[i], minus[i]),
                "-DI mismatch at index {}: baseline={} vs into={}",
                i,
                baseline.minus[i],
                minus[i]
            );
        }
        Ok(())
    }

    macro_rules! generate_all_di_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_di_tests!(
        check_di_partial_params,
        check_di_accuracy,
        check_di_with_zero_period,
        check_di_with_period_exceeding_data_length,
        check_di_very_small_data_set,
        check_di_with_slice_data_reinput,
        check_di_accuracy_nan_check,
        check_di_no_poison
    );

    #[cfg(test)]
    generate_all_di_tests!(check_di_property);
    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 = DiBatchBuilder::new().kernel(kernel).apply_candles(&c)?;

        let def = DiParams::default();
        let row = output.plus_for(&def).expect("default row missing");

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

        let scalar = DiInput::from_candles(&c, DiParams::default());
        let scalar_out = di_with_kernel(&scalar, Kernel::Scalar)?;
        let plus_tail = &row[row.len() - 5..];
        let scalar_tail = &scalar_out.plus[scalar_out.plus.len() - 5..];
        for (i, (&a, &b)) in plus_tail.iter().zip(scalar_tail.iter()).enumerate() {
            assert!(
                (a - b).abs() < 1e-8,
                "[{test}] batch/scalar plus mismatch idx={i}: {a} vs {b}"
            );
        }
        Ok(())
    }

    fn check_batch_period_range(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 = DiBatchBuilder::new()
            .kernel(kernel)
            .period_range(10, 20, 5)
            .apply_candles(&c)?;

        assert_eq!(output.rows, 3);
        assert_eq!(output.cols, c.close.len());

        let periods = [10, 15, 20];
        for (i, p) in periods.iter().enumerate() {
            let param = DiParams { period: Some(*p) };
            let plus = output.plus_for(&param).expect("plus missing");
            let minus = output.minus_for(&param).expect("minus missing");
            assert_eq!(plus.len(), c.close.len());
            assert_eq!(minus.len(), c.close.len());
        }
        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 test_configs = vec![
            (2, 10, 2),
            (5, 25, 5),
            (30, 60, 15),
            (2, 5, 1),
            (10, 10, 0),
            (14, 14, 0),
            (50, 50, 0),
            (7, 21, 7),
        ];

        for (cfg_idx, &(p_start, p_end, p_step)) in test_configs.iter().enumerate() {
            let output = DiBatchBuilder::new()
                .kernel(kernel)
                .period_range(p_start, p_end, p_step)
                .apply_candles(&c)?;

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

                let bits = val.to_bits();
                let row = idx / output.cols;
                let col = idx % output.cols;
                let combo = &output.combos[row];

                if bits == 0x11111111_11111111 {
                    panic!(
                        "[{}] Config {}: Found alloc_with_nan_prefix poison value {} (0x{:016X}) \
						at row {} col {} (flat index {}) in plus output with params: period={}",
                        test,
                        cfg_idx,
                        val,
                        bits,
                        row,
                        col,
                        idx,
                        combo.period.unwrap_or(14)
                    );
                }

                if bits == 0x22222222_22222222 {
                    panic!(
                        "[{}] Config {}: Found init_matrix_prefixes poison value {} (0x{:016X}) \
						at row {} col {} (flat index {}) in plus output with params: period={}",
                        test,
                        cfg_idx,
                        val,
                        bits,
                        row,
                        col,
                        idx,
                        combo.period.unwrap_or(14)
                    );
                }

                if bits == 0x33333333_33333333 {
                    panic!(
                        "[{}] Config {}: Found make_uninit_matrix poison value {} (0x{:016X}) \
						at row {} col {} (flat index {}) in plus output with params: period={}",
                        test,
                        cfg_idx,
                        val,
                        bits,
                        row,
                        col,
                        idx,
                        combo.period.unwrap_or(14)
                    );
                }
            }

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

                let bits = val.to_bits();
                let row = idx / output.cols;
                let col = idx % output.cols;
                let combo = &output.combos[row];

                if bits == 0x11111111_11111111 {
                    panic!(
                        "[{}] Config {}: Found alloc_with_nan_prefix poison value {} (0x{:016X}) \
						at row {} col {} (flat index {}) in minus output with params: period={}",
                        test,
                        cfg_idx,
                        val,
                        bits,
                        row,
                        col,
                        idx,
                        combo.period.unwrap_or(14)
                    );
                }

                if bits == 0x22222222_22222222 {
                    panic!(
                        "[{}] Config {}: Found init_matrix_prefixes poison value {} (0x{:016X}) \
						at row {} col {} (flat index {}) in minus output with params: period={}",
                        test,
                        cfg_idx,
                        val,
                        bits,
                        row,
                        col,
                        idx,
                        combo.period.unwrap_or(14)
                    );
                }

                if bits == 0x33333333_33333333 {
                    panic!(
                        "[{}] Config {}: Found make_uninit_matrix poison value {} (0x{:016X}) \
						at row {} col {} (flat index {}) in minus output with params: period={}",
                        test,
                        cfg_idx,
                        val,
                        bits,
                        row,
                        col,
                        idx,
                        combo.period.unwrap_or(14)
                    );
                }
            }
        }

        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_period_range);
    gen_batch_tests!(check_batch_no_poison);
}

#[cfg(feature = "python")]
#[pyfunction(name = "di")]
#[pyo3(signature = (high, low, close, period, kernel=None))]
pub fn di_py<'py>(
    py: Python<'py>,
    high: PyReadonlyArray1<'py, f64>,
    low: PyReadonlyArray1<'py, f64>,
    close: PyReadonlyArray1<'py, f64>,
    period: usize,
    kernel: Option<&str>,
) -> PyResult<(Bound<'py, PyArray1<f64>>, Bound<'py, PyArray1<f64>>)> {
    let high_slice = high.as_slice()?;
    let low_slice = low.as_slice()?;
    let close_slice = close.as_slice()?;
    let kern = validate_kernel(kernel, false)?;

    let params = DiParams {
        period: Some(period),
    };
    let input = DiInput::from_slices(high_slice, low_slice, close_slice, params);

    let (plus_vec, minus_vec) = py
        .allow_threads(|| di_with_kernel(&input, kern).map(|o| (o.plus, o.minus)))
        .map_err(|e| PyValueError::new_err(e.to_string()))?;

    Ok((plus_vec.into_pyarray(py), minus_vec.into_pyarray(py)))
}

#[cfg(feature = "python")]
#[pyclass(name = "DiStream")]
pub struct DiStreamPy {
    inner: DiStream,
}

#[cfg(feature = "python")]
#[pymethods]
impl DiStreamPy {
    #[new]
    pub fn new(period: usize) -> PyResult<Self> {
        let params = DiParams {
            period: Some(period),
        };
        let inner = DiStream::try_new(params).map_err(|e| PyValueError::new_err(e.to_string()))?;
        Ok(DiStreamPy { inner })
    }

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

#[cfg(feature = "python")]
#[pyfunction(name = "di_batch")]
#[pyo3(signature = (high, low, close, period_range, kernel=None))]
pub fn di_batch_py<'py>(
    py: Python<'py>,
    high: PyReadonlyArray1<'py, f64>,
    low: PyReadonlyArray1<'py, f64>,
    close: PyReadonlyArray1<'py, f64>,
    period_range: (usize, usize, usize),
    kernel: Option<&str>,
) -> PyResult<Bound<'py, PyDict>> {
    let high_slice = high.as_slice()?;
    let low_slice = low.as_slice()?;
    let close_slice = close.as_slice()?;
    let kern = validate_kernel(kernel, true)?;

    let sweep = DiBatchRange {
        period: period_range,
    };

    let combos = expand_grid(&sweep);
    let rows = combos.len();
    let cols = high_slice.len();
    let total = rows
        .checked_mul(cols)
        .ok_or_else(|| PyValueError::new_err("di: size overflow in rows*cols"))?;

    let out_plus = unsafe { PyArray1::<f64>::new(py, [total], false) };
    let out_minus = unsafe { PyArray1::<f64>::new(py, [total], false) };

    let pl = unsafe { out_plus.as_slice_mut()? };
    let mi = unsafe { out_minus.as_slice_mut()? };

    let first = (0..cols)
        .find(|&i| !(high_slice[i].is_nan() || low_slice[i].is_nan() || close_slice[i].is_nan()))
        .ok_or_else(|| PyValueError::new_err("di: All values are NaN"))?;
    let warm: Vec<usize> = combos
        .iter()
        .map(|p| first + p.period.unwrap() - 1)
        .collect();

    unsafe {
        let plus_mu = std::slice::from_raw_parts_mut(
            pl.as_mut_ptr() as *mut std::mem::MaybeUninit<f64>,
            total,
        );
        let minus_mu = std::slice::from_raw_parts_mut(
            mi.as_mut_ptr() as *mut std::mem::MaybeUninit<f64>,
            total,
        );
        init_matrix_prefixes(plus_mu, cols, &warm);
        init_matrix_prefixes(minus_mu, cols, &warm);
    }

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

            di_batch_inner_into(
                high_slice,
                low_slice,
                close_slice,
                &sweep,
                simd,
                true,
                pl,
                mi,
            )
        })
        .map_err(|e| PyValueError::new_err(e.to_string()))?;

    let dict = PyDict::new(py);
    dict.set_item("plus", out_plus.reshape((rows, cols))?)?;
    dict.set_item("minus", out_minus.reshape((rows, cols))?)?;
    dict.set_item(
        "periods",
        combos
            .iter()
            .map(|p| p.period.unwrap() as u64)
            .collect::<Vec<_>>()
            .into_pyarray(py),
    )?;

    Ok(dict)
}

#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::cuda_available;
#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::moving_averages::DeviceArrayF32;
#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::CudaDi;
#[cfg(all(feature = "python", feature = "cuda"))]
use crate::utilities::dlpack_cuda::export_f32_cuda_dlpack_2d;
#[cfg(all(feature = "python", feature = "cuda"))]
use cust::context::Context;
#[cfg(all(feature = "python", feature = "cuda"))]
use cust::memory::DeviceBuffer;
#[cfg(all(feature = "python", feature = "cuda"))]
use pyo3::prelude::*;
#[cfg(all(feature = "python", feature = "cuda"))]
use std::os::raw::c_void;
#[cfg(all(feature = "python", feature = "cuda"))]
use std::sync::Arc;

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyclass(module = "ta_indicators.cuda", unsendable)]
pub struct DeviceArrayF32DiPy {
    pub(crate) inner: DeviceArrayF32,
    pub(crate) ctx: Arc<Context>,
    pub(crate) device_id: u32,
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pymethods]
impl DeviceArrayF32DiPy {
    #[getter]
    fn __cuda_array_interface__<'py>(&self, py: Python<'py>) -> PyResult<Bound<'py, PyDict>> {
        let itemsize = std::mem::size_of::<f32>();
        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 * itemsize, itemsize))?;
        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.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<PyObject>,
        max_version: Option<PyObject>,
        dl_device: Option<PyObject>,
        copy: Option<PyObject>,
    ) -> PyResult<PyObject> {
        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 inner = std::mem::replace(
            &mut self.inner,
            DeviceArrayF32 {
                buf: dummy,
                rows: 0,
                cols: 0,
            },
        );

        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(all(feature = "python", feature = "cuda"))]
#[pyfunction(name = "di_cuda_batch_dev")]
#[pyo3(signature = (high_f32, low_f32, close_f32, period_range, device_id=0))]
pub fn di_cuda_batch_dev_py<'py>(
    py: Python<'py>,
    high_f32: numpy::PyReadonlyArray1<'py, f32>,
    low_f32: numpy::PyReadonlyArray1<'py, f32>,
    close_f32: numpy::PyReadonlyArray1<'py, f32>,
    period_range: (usize, usize, usize),
    device_id: usize,
) -> PyResult<Bound<'py, pyo3::types::PyDict>> {
    use numpy::IntoPyArray;
    if !cuda_available() {
        return Err(PyValueError::new_err("CUDA not available"));
    }
    let h = high_f32.as_slice()?;
    let l = low_f32.as_slice()?;
    let c = close_f32.as_slice()?;
    let sweep = DiBatchRange {
        period: period_range,
    };
    let (plus_dev, minus_dev, combos, ctx, dev_id) = py.allow_threads(|| {
        let cuda = CudaDi::new(device_id).map_err(|e| PyValueError::new_err(e.to_string()))?;
        let res = cuda
            .di_batch_dev(h, l, c, &sweep)
            .map_err(|e| PyValueError::new_err(e.to_string()))?;
        Ok::<_, pyo3::PyErr>((res.0, res.1, res.2, cuda.context_arc(), cuda.device_id()))
    })?;
    let dict = pyo3::types::PyDict::new(py);
    dict.set_item(
        "plus",
        Py::new(
            py,
            DeviceArrayF32DiPy {
                inner: plus_dev,
                ctx: ctx.clone(),
                device_id: dev_id,
            },
        )?,
    )?;
    dict.set_item(
        "minus",
        Py::new(
            py,
            DeviceArrayF32DiPy {
                inner: minus_dev,
                ctx,
                device_id: dev_id,
            },
        )?,
    )?;
    dict.set_item(
        "periods",
        combos
            .iter()
            .map(|p| p.period.unwrap() as u64)
            .collect::<Vec<_>>()
            .into_pyarray(py),
    )?;
    dict.set_item("rows", combos.len())?;
    dict.set_item("cols", h.len())?;
    Ok(dict)
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyfunction(name = "di_cuda_many_series_one_param_dev")]
#[pyo3(signature = (high_tm_f32, low_tm_f32, close_tm_f32, period, device_id=0))]
pub fn di_cuda_many_series_one_param_dev_py<'py>(
    py: Python<'py>,
    high_tm_f32: numpy::PyReadonlyArray2<'py, f32>,
    low_tm_f32: numpy::PyReadonlyArray2<'py, f32>,
    close_tm_f32: numpy::PyReadonlyArray2<'py, f32>,
    period: usize,
    device_id: usize,
) -> PyResult<Bound<'py, pyo3::types::PyDict>> {
    if !cuda_available() {
        return Err(PyValueError::new_err("CUDA not available"));
    }
    let h_shape = high_tm_f32.shape();
    let l_shape = low_tm_f32.shape();
    let c_shape = close_tm_f32.shape();
    if h_shape != l_shape || l_shape != c_shape || h_shape.len() != 2 {
        return Err(PyValueError::new_err(
            "expected three 2D arrays of same shape",
        ));
    }
    let rows = h_shape[0];
    let cols = h_shape[1];
    let h = high_tm_f32.as_slice()?;
    let l = low_tm_f32.as_slice()?;
    let c = close_tm_f32.as_slice()?;
    let (pair, ctx, dev_id) = py.allow_threads(|| {
        let cuda = CudaDi::new(device_id).map_err(|e| PyValueError::new_err(e.to_string()))?;
        let p = cuda
            .di_many_series_one_param_time_major_dev(h, l, c, cols, rows, period)
            .map_err(|e| PyValueError::new_err(e.to_string()))?;
        Ok::<_, pyo3::PyErr>((p, cuda.context_arc(), cuda.device_id()))
    })?;
    let dict = pyo3::types::PyDict::new(py);
    dict.set_item(
        "plus",
        Py::new(
            py,
            DeviceArrayF32DiPy {
                inner: pair.plus,
                ctx: ctx.clone(),
                device_id: dev_id,
            },
        )?,
    )?;
    dict.set_item(
        "minus",
        Py::new(
            py,
            DeviceArrayF32DiPy {
                inner: pair.minus,
                ctx,
                device_id: dev_id,
            },
        )?,
    )?;
    dict.set_item("rows", rows)?;
    dict.set_item("cols", cols)?;
    dict.set_item("period", period)?;
    Ok(dict)
}

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

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen(js_name = di)]
pub fn di_js(high: &[f64], low: &[f64], close: &[f64], period: usize) -> Result<JsValue, JsValue> {
    let params = DiParams {
        period: Some(period),
    };
    let input = DiInput::from_slices(high, low, close, params);
    let out = di_with_kernel(&input, crate::utilities::enums::Kernel::Auto)
        .map_err(|e| JsValue::from_str(&e.to_string()))?;
    let cols = high.len();
    let total = cols
        .checked_mul(2)
        .ok_or_else(|| JsValue::from_str("di_js: size overflow"))?;
    let mut values = Vec::with_capacity(total);
    values.extend_from_slice(&out.plus);
    values.extend_from_slice(&out.minus);
    let result = DiJsResult {
        values,
        rows: 2,
        cols,
    };
    serde_wasm_bindgen::to_value(&result).map_err(|e| JsValue::from_str(&e.to_string()))
}

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

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn di_into(
    high_ptr: *const f64,
    low_ptr: *const f64,
    close_ptr: *const f64,
    out_ptr: *mut f64,
    len: usize,
    period: usize,
) -> Result<(), JsValue> {
    if high_ptr.is_null() || low_ptr.is_null() || close_ptr.is_null() || out_ptr.is_null() {
        return Err(JsValue::from_str("null pointer to di_into"));
    }
    unsafe {
        let h = std::slice::from_raw_parts(high_ptr, len);
        let l = std::slice::from_raw_parts(low_ptr, len);
        let c = std::slice::from_raw_parts(close_ptr, len);
        let params = DiParams {
            period: Some(period),
        };
        let input = DiInput::from_slices(h, l, c, params);

        let DiOutput { plus, minus } =
            di_with_kernel(&input, crate::utilities::enums::Kernel::Auto)
                .map_err(|e| JsValue::from_str(&e.to_string()))?;

        let total = len
            .checked_mul(2)
            .ok_or_else(|| JsValue::from_str("di_into: size overflow"))?;
        let out = std::slice::from_raw_parts_mut(out_ptr, total);
        out[..len].copy_from_slice(&plus);
        out[len..total].copy_from_slice(&minus);
        Ok(())
    }
}

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

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

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen(js_name = di_batch)]
pub fn di_batch_unified_js(
    high: &[f64],
    low: &[f64],
    close: &[f64],
    config: JsValue,
) -> Result<JsValue, JsValue> {
    let cfg: DiBatchConfig = serde_wasm_bindgen::from_value(config)
        .map_err(|e| JsValue::from_str(&format!("Invalid config: {}", e)))?;
    let sweep = DiBatchRange {
        period: cfg.period_range,
    };
    let output = di_batch_slice(
        high,
        low,
        close,
        &sweep,
        crate::utilities::enums::Kernel::Scalar,
    )
    .map_err(|e| JsValue::from_str(&e.to_string()))?;
    let rows = output
        .rows
        .checked_mul(2)
        .ok_or_else(|| JsValue::from_str("di_batch_unified_js: rows overflow"))?;
    let cols = output.cols;

    let total = rows
        .checked_mul(cols)
        .ok_or_else(|| JsValue::from_str("di_batch_unified_js: size overflow"))?;
    let mut values = Vec::with_capacity(total);

    for combo_idx in 0..output.rows {
        let start = combo_idx * cols;
        values.extend_from_slice(&output.plus[start..start + cols]);
        values.extend_from_slice(&output.minus[start..start + cols]);
    }

    let js = DiBatchJsOutput {
        values,
        rows,
        cols,
        periods: output
            .combos
            .iter()
            .map(|p| p.period.unwrap())
            .collect::<Vec<_>>(),
    };
    serde_wasm_bindgen::to_value(&js)
        .map_err(|e| JsValue::from_str(&format!("Serialization error: {}", e)))
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn di_batch_into(
    high_ptr: *const f64,
    low_ptr: *const f64,
    close_ptr: *const f64,
    plus_ptr: *mut f64,
    minus_ptr: *mut f64,
    len: usize,
    period_start: usize,
    period_end: usize,
    period_step: usize,
) -> Result<usize, JsValue> {
    if high_ptr.is_null()
        || low_ptr.is_null()
        || close_ptr.is_null()
        || plus_ptr.is_null()
        || minus_ptr.is_null()
    {
        return Err(JsValue::from_str("Null pointer provided"));
    }

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

        let sweep = DiBatchRange {
            period: (period_start, period_end, period_step),
        };

        let combos = expand_grid(&sweep);
        let rows = combos.len();
        let total_size = rows
            .checked_mul(len)
            .ok_or_else(|| JsValue::from_str("di_batch_into: size overflow"))?;

        let out_plus = std::slice::from_raw_parts_mut(plus_ptr, total_size);
        let out_minus = std::slice::from_raw_parts_mut(minus_ptr, total_size);

        di_batch_inner_into(
            high,
            low,
            close,
            &sweep,
            Kernel::Auto,
            false,
            out_plus,
            out_minus,
        )
        .map_err(|e| JsValue::from_str(&e.to_string()))?;

        Ok(rows)
    }
}