vector_ta/indicators/moving_averages/

jsa.rs

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 std::mem::MaybeUninit;
use thiserror::Error;

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

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

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

#[derive(Debug, Clone)]
pub struct JsaParams {
    pub period: Option<usize>,
}

impl Default for JsaParams {
    fn default() -> Self {
        Self { period: Some(30) }
    }
}

#[derive(Debug, Clone)]
pub struct JsaInput<'a> {
    pub data: JsaData<'a>,
    pub params: JsaParams,
}

impl<'a> JsaInput<'a> {
    #[inline(always)]
    pub fn from_candles(c: &'a Candles, s: &'a str, p: JsaParams) -> Self {
        Self {
            data: JsaData::Candles {
                candles: c,
                source: s,
            },
            params: p,
        }
    }
    #[inline(always)]
    pub fn from_slice(sl: &'a [f64], p: JsaParams) -> Self {
        Self {
            data: JsaData::Slice(sl),
            params: p,
        }
    }
    #[inline(always)]
    pub fn with_default_candles(c: &'a Candles) -> Self {
        Self::from_candles(c, "close", JsaParams::default())
    }
    #[inline(always)]
    pub fn get_period(&self) -> usize {
        self.params.period.unwrap_or(30)
    }
}

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

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

impl JsaBuilder {
    #[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, c: &Candles) -> Result<JsaOutput, JsaError> {
        let p = JsaParams {
            period: self.period,
        };
        let i = JsaInput::from_candles(c, "close", p);
        jsa_with_kernel(&i, self.kernel)
    }
    #[inline(always)]
    pub fn apply_slice(self, d: &[f64]) -> Result<JsaOutput, JsaError> {
        let p = JsaParams {
            period: self.period,
        };
        let i = JsaInput::from_slice(d, p);
        jsa_with_kernel(&i, self.kernel)
    }
    #[inline(always)]
    pub fn into_stream(self) -> Result<JsaStream, JsaError> {
        let p = JsaParams {
            period: self.period,
        };
        JsaStream::try_new(p)
    }
}

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

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

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

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

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

    #[error("jsa: invalid kernel for batch op: {kernel:?}")]
    InvalidKernel { kernel: Kernel },

    #[error("jsa: invalid range expansion: start={start}, end={end}, step={step}")]
    InvalidRange {
        start: usize,
        end: usize,
        step: usize,
    },

    #[error("jsa: arithmetic overflow while computing sizes")]
    ArithmeticOverflow,

    #[error("jsa: invalid kernel used for batch op: {kernel:?}")]
    InvalidKernelForBatch { kernel: Kernel },
}

#[inline]
pub fn jsa(input: &JsaInput) -> Result<JsaOutput, JsaError> {
    jsa_with_kernel(input, Kernel::Auto)
}

#[inline(always)]
fn jsa_compute_into(data: &[f64], period: usize, first: usize, k: Kernel, out: &mut [f64]) {
    unsafe {
        match k {
            Kernel::Scalar | Kernel::ScalarBatch => jsa_scalar(data, period, first, out),
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx2 | Kernel::Avx2Batch => jsa_avx2(data, period, first, out),
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx512 | Kernel::Avx512Batch => jsa_avx512(data, period, first, out),
            _ => unreachable!(),
        }
    }
}

pub fn jsa_with_kernel(input: &JsaInput, kernel: Kernel) -> Result<JsaOutput, JsaError> {
    let data: &[f64] = match &input.data {
        JsaData::Candles { candles, source } => source_type(candles, source),
        JsaData::Slice(sl) => sl,
    };

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

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

    if period == 0 || period > len {
        return Err(JsaError::InvalidPeriod {
            period,
            data_len: len,
        });
    }
    if (len - first) < period {
        return Err(JsaError::NotEnoughValidData {
            needed: period,
            valid: len - first,
        });
    }

    let warm = first
        .checked_add(period)
        .ok_or(JsaError::ArithmeticOverflow)?;
    let mut out = alloc_with_nan_prefix(len, warm);
    let chosen = match kernel {
        Kernel::Auto => Kernel::Scalar,
        k => k,
    };
    jsa_compute_into(data, period, first, chosen, &mut out);
    Ok(JsaOutput { values: out })
}

#[inline]
#[inline]
pub fn jsa_into(input: &JsaInput, out: &mut [f64]) -> Result<(), JsaError> {
    jsa_with_kernel_into(input, Kernel::Auto, out)
}

#[inline]
pub fn jsa_into_slice(dst: &mut [f64], input: &JsaInput, kern: Kernel) -> Result<(), JsaError> {
    jsa_with_kernel_into(input, kern, dst)
}

pub fn jsa_with_kernel_into(
    input: &JsaInput,
    kernel: Kernel,
    out: &mut [f64],
) -> Result<(), JsaError> {
    let data: &[f64] = match &input.data {
        JsaData::Candles { candles, source } => source_type(candles, source),
        JsaData::Slice(sl) => sl,
    };

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

    let len = data.len();

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

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

    if period == 0 || period > len {
        return Err(JsaError::InvalidPeriod {
            period,
            data_len: len,
        });
    }
    if (len - first) < period {
        return Err(JsaError::NotEnoughValidData {
            needed: period,
            valid: len - first,
        });
    }

    let warm = first + period;

    out[..warm].fill(f64::NAN);

    let chosen = match kernel {
        Kernel::Auto => Kernel::Scalar,
        k => k,
    };
    jsa_compute_into(data, period, first, chosen, out);
    Ok(())
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline]
pub fn jsa_avx512(data: &[f64], period: usize, first_valid: usize, out: &mut [f64]) {
    if period <= 32 {
        unsafe { jsa_avx512_short(data, period, first_valid, out) }
    } else {
        unsafe { jsa_avx512_long(data, period, first_valid, out) }
    }
}

#[inline]
pub fn jsa_scalar(data: &[f64], period: usize, first_val: usize, out: &mut [f64]) {
    for i in (first_val + period)..data.len() {
        out[i] = (data[i] + data[i - period]) * 0.5;
    }
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn jsa_avx2(data: &[f64], period: usize, first_val: usize, out: &mut [f64]) {
    use core::arch::x86_64::*;
    let len = data.len();
    let start = first_val + period;
    if start >= len {
        return;
    }

    let dp = data.as_ptr();
    let op = out.as_mut_ptr();

    let mut p_cur = dp.add(start);
    let mut p_past = dp.add(start - period);
    let mut p_out = op.add(start);
    let end = op.add(len);

    let half = _mm256_set1_pd(0.5);

    while p_out.add(4) <= end {
        let x = _mm256_loadu_pd(p_cur);
        let y = _mm256_loadu_pd(p_past);
        let s = _mm256_add_pd(x, y);
        let a = _mm256_mul_pd(s, half);
        _mm256_storeu_pd(p_out, a);
        p_cur = p_cur.add(4);
        p_past = p_past.add(4);
        p_out = p_out.add(4);
    }

    while p_out < end {
        *p_out = (*p_cur + *p_past) * 0.5;
        p_cur = p_cur.add(1);
        p_past = p_past.add(1);
        p_out = p_out.add(1);
    }
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn jsa_avx512_short(data: &[f64], period: usize, first_val: usize, out: &mut [f64]) {
    use core::arch::x86_64::*;
    let len = data.len();
    let start = first_val + period;
    if start >= len {
        return;
    }

    let dp = data.as_ptr();
    let op = out.as_mut_ptr();

    let mut p_cur = dp.add(start);
    let mut p_past = dp.add(start - period);
    let mut p_out = op.add(start);
    let end = op.add(len);

    let half = _mm512_set1_pd(0.5);

    while p_out.add(8) <= end {
        let x = _mm512_loadu_pd(p_cur);
        let y = _mm512_loadu_pd(p_past);
        let s = _mm512_add_pd(x, y);
        let a = _mm512_mul_pd(s, half);
        _mm512_storeu_pd(p_out, a);
        p_cur = p_cur.add(8);
        p_past = p_past.add(8);
        p_out = p_out.add(8);
    }

    while p_out < end {
        *p_out = (*p_cur + *p_past) * 0.5;
        p_cur = p_cur.add(1);
        p_past = p_past.add(1);
        p_out = p_out.add(1);
    }
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
pub unsafe fn jsa_avx512_long(data: &[f64], period: usize, first_val: usize, out: &mut [f64]) {
    use core::arch::x86_64::*;
    let len = data.len();
    let start = first_val + period;
    if start >= len {
        return;
    }

    let dp = data.as_ptr();
    let op = out.as_mut_ptr();

    let mut p_cur = dp.add(start);
    let mut p_past = dp.add(start - period);
    let mut p_out = op.add(start);
    let end = op.add(len);

    let half = _mm512_set1_pd(0.5);

    while p_out.add(8) <= end {
        let x = _mm512_loadu_pd(p_cur);
        let y = _mm512_loadu_pd(p_past);
        let s = _mm512_add_pd(x, y);
        let a = _mm512_mul_pd(s, half);
        _mm512_storeu_pd(p_out, a);
        p_cur = p_cur.add(8);
        p_past = p_past.add(8);
        p_out = p_out.add(8);
    }
    while p_out < end {
        *p_out = (*p_cur + *p_past) * 0.5;
        p_cur = p_cur.add(1);
        p_past = p_past.add(1);
        p_out = p_out.add(1);
    }
}

#[derive(Debug, Clone)]
pub struct JsaStream {
    period: usize,
    buffer: Vec<f64>,
    head: usize,
    filled: bool,
}

impl JsaStream {
    pub fn try_new(params: JsaParams) -> Result<Self, JsaError> {
        let period = params.period.unwrap_or(30);
        if period == 0 {
            return Err(JsaError::InvalidPeriod {
                period,
                data_len: 0,
            });
        }
        Ok(Self {
            period,
            buffer: vec![f64::NAN; period],
            head: 0,
            filled: false,
        })
    }

    #[inline(always)]
    pub fn update(&mut self, value: f64) -> Option<f64> {
        let out = if self.filled {
            let past = self.buffer[self.head];
            Some((value + past) * 0.5)
        } else {
            None
        };

        self.buffer[self.head] = value;

        let next = self.head + 1;
        if next == self.period {
            self.head = 0;
            if !self.filled {
                self.filled = true;
            }
        } else {
            self.head = next;
        }

        out
    }
}

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

impl Default for JsaBatchRange {
    fn default() -> Self {
        Self {
            period: (30, 279, 1),
        }
    }
}

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

impl JsaBatchBuilder {
    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
    }
    pub fn apply_slice(self, data: &[f64]) -> Result<JsaBatchOutput, JsaError> {
        jsa_batch_with_kernel(data, &self.range, self.kernel)
    }
    pub fn with_default_slice(data: &[f64], k: Kernel) -> Result<JsaBatchOutput, JsaError> {
        JsaBatchBuilder::new().kernel(k).apply_slice(data)
    }
    pub fn apply_candles(self, c: &Candles, src: &str) -> Result<JsaBatchOutput, JsaError> {
        let slice = source_type(c, src);
        self.apply_slice(slice)
    }
    pub fn with_default_candles(c: &Candles) -> Result<JsaBatchOutput, JsaError> {
        JsaBatchBuilder::new()
            .kernel(Kernel::Auto)
            .apply_candles(c, "close")
    }
}

pub fn jsa_batch_with_kernel(
    data: &[f64],
    sweep: &JsaBatchRange,
    k: Kernel,
) -> Result<JsaBatchOutput, JsaError> {
    let kernel = match k {
        Kernel::Auto => detect_best_batch_kernel(),
        other if other.is_batch() => other,
        other => return Err(JsaError::InvalidKernelForBatch { kernel: other }),
    };

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

#[derive(Clone, Debug)]
pub struct JsaBatchOutput {
    pub values: Vec<f64>,
    pub combos: Vec<JsaParams>,
    pub rows: usize,
    pub cols: usize,
}
impl JsaBatchOutput {
    pub fn row_for_params(&self, p: &JsaParams) -> Option<usize> {
        self.combos
            .iter()
            .position(|c| c.period.unwrap_or(30) == p.period.unwrap_or(30))
    }
    pub fn values_for(&self, p: &JsaParams) -> 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: &JsaBatchRange) -> Result<Vec<JsaParams>, JsaError> {
    fn axis_usize((start, end, step): (usize, usize, usize)) -> Result<Vec<usize>, JsaError> {
        if step == 0 || start == end {
            return Ok(vec![start]);
        }
        let mut v = Vec::new();
        if start < end {
            let mut cur = start;
            while cur <= end {
                v.push(cur);

                match cur.checked_add(step) {
                    Some(next) => cur = next,
                    None => break,
                }
            }
        } else {
            let mut cur = start;
            while cur >= end {
                v.push(cur);

                if cur < end + step {
                    break;
                }
                cur -= step;
                if cur == usize::MAX {
                    break;
                }
            }
        }
        if v.is_empty() {
            return Err(JsaError::InvalidRange { start, end, step });
        }
        Ok(v)
    }
    let periods = axis_usize(r.period)?;
    let mut out = Vec::with_capacity(periods.len());
    for &p in &periods {
        out.push(JsaParams { period: Some(p) });
    }
    Ok(out)
}

#[inline(always)]
pub fn jsa_batch_slice(
    data: &[f64],
    sweep: &JsaBatchRange,
    kern: Kernel,
) -> Result<JsaBatchOutput, JsaError> {
    jsa_batch_inner(data, sweep, kern, false)
}

#[inline(always)]
pub fn jsa_batch_par_slice(
    data: &[f64],
    sweep: &JsaBatchRange,
    kern: Kernel,
) -> Result<JsaBatchOutput, JsaError> {
    jsa_batch_inner(data, sweep, kern, true)
}

#[inline(always)]
fn jsa_batch_inner(
    data: &[f64],
    sweep: &JsaBatchRange,
    kern: Kernel,
    parallel: bool,
) -> Result<JsaBatchOutput, JsaError> {
    if data.is_empty() {
        return Err(JsaError::EmptyInputData);
    }

    let combos = expand_grid(sweep)?;
    let first = data
        .iter()
        .position(|x| !x.is_nan())
        .ok_or(JsaError::AllValuesNaN)?;
    let max_p = combos.iter().map(|c| c.period.unwrap()).max().unwrap();
    if data.len() - first < max_p {
        return Err(JsaError::NotEnoughValidData {
            needed: max_p,
            valid: data.len() - first,
        });
    }
    let rows = combos.len();
    let cols = data.len();

    let _total = rows.checked_mul(cols).ok_or(JsaError::ArithmeticOverflow)?;
    let mut warm: Vec<usize> = Vec::with_capacity(rows);
    for c in &combos {
        let p = c.period.unwrap();
        let w = first.checked_add(p).ok_or(JsaError::ArithmeticOverflow)?;
        warm.push(w);
    }

    let mut raw = make_uninit_matrix(rows, cols);

    init_matrix_prefixes(&mut raw, cols, &warm);

    let actual_kern = match kern {
        Kernel::Auto => detect_best_batch_kernel(),
        k => k,
    };

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

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

        match actual_kern {
            Kernel::ScalarBatch | Kernel::Scalar => jsa_row_scalar(data, first, period, out_row),
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx2Batch | Kernel::Avx2 => jsa_row_avx2(data, first, period, out_row),
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx512Batch | Kernel::Avx512 => jsa_row_avx512(data, first, period, out_row),
            _ => unreachable!(),
        }
    };

    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);
        }
    }

    use core::mem::ManuallyDrop;

    let mut buf_guard = ManuallyDrop::new(raw);
    let values = unsafe {
        Vec::from_raw_parts(
            buf_guard.as_mut_ptr() as *mut f64,
            buf_guard.len(),
            buf_guard.capacity(),
        )
    };

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

#[inline(always)]
fn jsa_batch_inner_into(
    data: &[f64],
    sweep: &JsaBatchRange,
    kern: Kernel,
    parallel: bool,
    out: &mut [f64],
) -> Result<(Vec<JsaParams>, usize, usize), JsaError> {
    if data.is_empty() {
        return Err(JsaError::EmptyInputData);
    }

    let combos = expand_grid(sweep)?;
    let first = data
        .iter()
        .position(|x| !x.is_nan())
        .ok_or(JsaError::AllValuesNaN)?;
    let max_p = combos.iter().map(|c| c.period.unwrap()).max().unwrap();
    if data.len() - first < max_p {
        return Err(JsaError::NotEnoughValidData {
            needed: max_p,
            valid: data.len() - first,
        });
    }
    let rows = combos.len();
    let cols = data.len();

    let expected = rows.checked_mul(cols).ok_or(JsaError::ArithmeticOverflow)?;
    if out.len() != expected {
        return Err(JsaError::OutputLengthMismatch {
            expected,
            got: out.len(),
        });
    }
    let mut warm: Vec<usize> = Vec::with_capacity(rows);
    for c in &combos {
        let p = c.period.unwrap();
        let w = first.checked_add(p).ok_or(JsaError::ArithmeticOverflow)?;
        warm.push(w);
    }

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

    let actual_kern = match kern {
        Kernel::Auto => detect_best_batch_kernel(),
        k => k,
    };

    let do_row = |row: usize, out_row: &mut [f64]| unsafe {
        let period = combos[row].period.unwrap();

        match actual_kern {
            Kernel::ScalarBatch | Kernel::Scalar => jsa_row_scalar(data, first, period, out_row),
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx2Batch | Kernel::Avx2 => jsa_row_avx2(data, first, period, out_row),
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx512Batch | Kernel::Avx512 => jsa_row_avx512(data, first, period, out_row),
            _ => unreachable!(),
        }
    };

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

    Ok((combos, rows, cols))
}

#[inline(always)]
unsafe fn jsa_row_scalar(data: &[f64], first: usize, period: usize, out: &mut [f64]) {
    for i in (first + period)..data.len() {
        out[i] = (data[i] + data[i - period]) * 0.5;
    }
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
unsafe fn jsa_row_avx2(data: &[f64], first: usize, period: usize, out: &mut [f64]) {
    use core::arch::x86_64::*;
    let len = data.len();
    let start = first + period;
    if start >= len {
        return;
    }

    let dp = data.as_ptr();
    let op = out.as_mut_ptr();

    let mut p_cur = dp.add(start);
    let mut p_past = dp.add(start - period);
    let mut p_out = op.add(start);
    let end = op.add(len);

    let half = _mm256_set1_pd(0.5);

    while p_out.add(4) <= end {
        let x = _mm256_loadu_pd(p_cur);
        let y = _mm256_loadu_pd(p_past);
        let s = _mm256_add_pd(x, y);
        let a = _mm256_mul_pd(s, half);
        _mm256_storeu_pd(p_out, a);
        p_cur = p_cur.add(4);
        p_past = p_past.add(4);
        p_out = p_out.add(4);
    }
    while p_out < end {
        *p_out = (*p_cur + *p_past) * 0.5;
        p_cur = p_cur.add(1);
        p_past = p_past.add(1);
        p_out = p_out.add(1);
    }
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
unsafe fn jsa_row_avx512(data: &[f64], first: usize, period: usize, out: &mut [f64]) {
    if period <= 32 {
        jsa_row_avx512_short(data, first, period, out);
    } else {
        jsa_row_avx512_long(data, first, period, out);
    }
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
unsafe fn jsa_row_avx512_short(data: &[f64], first: usize, period: usize, out: &mut [f64]) {
    use core::arch::x86_64::*;
    let len = data.len();
    let start = first + period;
    if start >= len {
        return;
    }

    let dp = data.as_ptr();
    let op = out.as_mut_ptr();

    let mut p_cur = dp.add(start);
    let mut p_past = dp.add(start - period);
    let mut p_out = op.add(start);
    let end = op.add(len);

    let half = _mm512_set1_pd(0.5);

    while p_out.add(8) <= end {
        let x = _mm512_loadu_pd(p_cur);
        let y = _mm512_loadu_pd(p_past);
        let s = _mm512_add_pd(x, y);
        let a = _mm512_mul_pd(s, half);
        _mm512_storeu_pd(p_out, a);
        p_cur = p_cur.add(8);
        p_past = p_past.add(8);
        p_out = p_out.add(8);
    }
    while p_out < end {
        *p_out = (*p_cur + *p_past) * 0.5;
        p_cur = p_cur.add(1);
        p_past = p_past.add(1);
        p_out = p_out.add(1);
    }
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[inline(always)]
unsafe fn jsa_row_avx512_long(data: &[f64], first: usize, period: usize, out: &mut [f64]) {
    use core::arch::x86_64::*;
    let len = data.len();
    let start = first + period;
    if start >= len {
        return;
    }

    let dp = data.as_ptr();
    let op = out.as_mut_ptr();

    let mut p_cur = dp.add(start);
    let mut p_past = dp.add(start - period);
    let mut p_out = op.add(start);
    let end = op.add(len);

    let half = _mm512_set1_pd(0.5);

    while p_out.add(8) <= end {
        let x = _mm512_loadu_pd(p_cur);
        let y = _mm512_loadu_pd(p_past);
        let s = _mm512_add_pd(x, y);
        let a = _mm512_mul_pd(s, half);
        _mm512_storeu_pd(p_out, a);
        p_cur = p_cur.add(8);
        p_past = p_past.add(8);
        p_out = p_out.add(8);
    }
    while p_out < end {
        *p_out = (*p_cur + *p_past) * 0.5;
        p_cur = p_cur.add(1);
        p_past = p_past.add(1);
        p_out = p_out.add(1);
    }
}

#[cfg(feature = "python")]
use numpy::PyUntypedArrayMethods;
#[cfg(feature = "python")]
use numpy::{IntoPyArray, PyArray1, PyArrayMethods, PyReadonlyArray1, PyReadonlyArray2};
#[cfg(feature = "python")]
use pyo3::exceptions::PyValueError;
#[cfg(feature = "python")]
use pyo3::prelude::*;
#[cfg(feature = "python")]
use pyo3::types::PyDict;

#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::moving_averages::jsa_wrapper::JsaDeviceHandle;
#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::{cuda_available, moving_averages::CudaJsa};
#[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 std::sync::Arc;

#[cfg(feature = "python")]
#[pyfunction]
#[pyo3(name = "jsa")]
#[pyo3(signature = (data, period, kernel=None))]
pub fn jsa_py<'py>(
    py: Python<'py>,
    data: PyReadonlyArray1<'py, f64>,
    period: usize,
    kernel: Option<&str>,
) -> PyResult<Bound<'py, PyArray1<f64>>> {
    use numpy::PyArrayMethods;

    let kern = validate_kernel(kernel, false)?;

    if data.is_c_contiguous() {
        let slice_in = data.as_slice()?;

        let out_arr = unsafe { PyArray1::<f64>::new(py, [slice_in.len()], false) };
        let slice_out = unsafe { out_arr.as_slice_mut()? };
        let params = JsaParams {
            period: Some(period),
        };
        let input = JsaInput::from_slice(slice_in, params);
        py.allow_threads(|| jsa_with_kernel_into(&input, kern, slice_out))
            .map_err(|e| PyValueError::new_err(e.to_string()))?;
        Ok(out_arr)
    } else {
        let owned = data.as_array().to_owned();
        let slice_in = owned.as_slice().expect("owned array should be contiguous");
        let params = JsaParams {
            period: Some(period),
        };
        let input = JsaInput::from_slice(slice_in, params);
        let mut buf = vec![f64::NAN; slice_in.len()];
        py.allow_threads(|| jsa_with_kernel_into(&input, kern, &mut buf))
            .map_err(|e| PyValueError::new_err(e.to_string()))?;
        Ok(PyArray1::from_vec(py, buf))
    }
}

#[cfg(feature = "python")]
#[pyfunction]
#[pyo3(name = "jsa_batch")]
#[pyo3(signature = (data, period_start, period_end, period_step, kernel=None))]
pub fn jsa_batch_py<'py>(
    py: Python<'py>,
    data: PyReadonlyArray1<'py, f64>,
    period_start: usize,
    period_end: usize,
    period_step: usize,
    kernel: Option<&str>,
) -> PyResult<Bound<'py, PyDict>> {
    use numpy::{IntoPyArray, PyArray1, PyArrayMethods};
    use pyo3::types::PyDict;

    let slice_in = data.as_slice()?;
    let kern = validate_kernel(kernel, true)?;
    let sweep = JsaBatchRange {
        period: (period_start, period_end, period_step),
    };

    let combos = expand_grid(&sweep).map_err(|e| PyValueError::new_err(e.to_string()))?;
    let rows = combos.len();
    let cols = slice_in.len();
    let total = rows
        .checked_mul(cols)
        .ok_or_else(|| PyValueError::new_err("size overflow"))?;

    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,
                _ => kernel,
            };

            jsa_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),
    )?;

    Ok(dict)
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyfunction(name = "jsa_cuda_batch_dev")]
#[pyo3(signature = (data_f32, period_range=(30, 30, 0), device_id=0))]
pub fn jsa_cuda_batch_dev_py(
    py: Python<'_>,
    data_f32: PyReadonlyArray1<'_, f32>,
    period_range: (usize, usize, usize),
    device_id: usize,
) -> PyResult<JsaDeviceArrayF32Py> {
    if !cuda_available() {
        return Err(PyValueError::new_err("CUDA not available"));
    }

    let slice_in = data_f32.as_slice()?;
    let sweep = JsaBatchRange {
        period: period_range,
    };

    use crate::cuda::moving_averages::CudaJsaError;
    let handle: JsaDeviceHandle = py
        .allow_threads(|| -> Result<_, CudaJsaError> {
            let cuda = CudaJsa::new(device_id)?;
            cuda.jsa_batch_dev_handle(slice_in, &sweep)
        })
        .map_err(|e| PyValueError::new_err(e.to_string()))?;

    Ok(JsaDeviceArrayF32Py::from_handle_rust(handle))
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyfunction(name = "jsa_cuda_many_series_one_param_dev")]
#[pyo3(signature = (data_tm_f32, period, device_id=0))]
pub fn jsa_cuda_many_series_one_param_dev_py(
    py: Python<'_>,
    data_tm_f32: PyReadonlyArray2<'_, f32>,
    period: usize,
    device_id: usize,
) -> PyResult<JsaDeviceArrayF32Py> {
    if !cuda_available() {
        return Err(PyValueError::new_err("CUDA not available"));
    }
    if period == 0 {
        return Err(PyValueError::new_err("period must be positive"));
    }

    let flat = data_tm_f32.as_slice()?;
    let shape = data_tm_f32.shape();
    if shape.len() != 2 {
        return Err(PyValueError::new_err("expected a 2D array"));
    }
    let series_len = shape[0];
    let num_series = shape[1];
    let params = JsaParams {
        period: Some(period),
    };

    use crate::cuda::moving_averages::CudaJsaError;
    let handle: JsaDeviceHandle = py
        .allow_threads(|| -> Result<_, CudaJsaError> {
            let cuda = CudaJsa::new(device_id)?;
            cuda.jsa_many_series_one_param_time_major_dev_handle(
                flat, num_series, series_len, &params,
            )
        })
        .map_err(|e| PyValueError::new_err(e.to_string()))?;

    Ok(JsaDeviceArrayF32Py::from_handle_rust(handle))
}

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

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

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

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
use serde::{Deserialize, Serialize};
#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
use wasm_bindgen::prelude::*;

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn jsa_js(data: &[f64], period: usize) -> Result<Vec<f64>, JsValue> {
    let params = JsaParams {
        period: Some(period),
    };
    let input = JsaInput::from_slice(data, params);

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

    jsa_into(&input, &mut output).map_err(|e| JsValue::from_str(&e.to_string()))?;

    Ok(output)
}

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

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

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

    let sweep = JsaBatchRange {
        period: config.period_range,
    };

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

    let js_output = JsaBatchJsOutput {
        values: output.values,
        periods: output
            .combos
            .iter()
            .map(|p| p.period.unwrap_or(30))
            .collect(),
        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(js_name = jsa_batch_simple)]
pub fn jsa_batch_simple(
    data: &[f64],
    period_start: usize,
    period_end: usize,
    period_step: usize,
) -> Result<Vec<f64>, JsValue> {
    let sweep = JsaBatchRange {
        period: (period_start, period_end, period_step),
    };

    jsa_batch_inner(data, &sweep, Kernel::Auto, false)
        .map(|output| output.values)
        .map_err(|e| JsValue::from_str(&e.to_string()))
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn jsa_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 jsa_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(js_name = jsa_into)]
pub fn jsa_into_wasm(
    in_ptr: *const f64,
    out_ptr: *mut f64,
    len: usize,
    period: usize,
) -> Result<(), JsValue> {
    if in_ptr.is_null() || out_ptr.is_null() {
        return Err(JsValue::from_str("null pointer passed to jsa_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 input = JsaInput::from_slice(
            data,
            JsaParams {
                period: Some(period),
            },
        );
        if in_ptr == out_ptr {
            let mut temp = vec![0.0; len];
            jsa_into_slice(&mut temp, &input, Kernel::Auto)
                .map_err(|e| JsValue::from_str(&e.to_string()))?;
            std::slice::from_raw_parts_mut(out_ptr, len).copy_from_slice(&temp);
        } else {
            jsa_into_slice(
                std::slice::from_raw_parts_mut(out_ptr, len),
                &input,
                Kernel::Auto,
            )
            .map_err(|e| JsValue::from_str(&e.to_string()))?;
        }
        Ok(())
    }
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
#[deprecated(note = "use jsa_into")]
pub fn jsa_fast(
    in_ptr: *const f64,
    out_ptr: *mut f64,
    len: usize,
    period: usize,
) -> Result<(), JsValue> {
    jsa_into_wasm(in_ptr, out_ptr, len, period)
}

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

    unsafe {
        let data = std::slice::from_raw_parts(in_ptr, len);
        let sweep = JsaBatchRange {
            period: (period_start, period_end, period_step),
        };

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

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

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

        Ok(rows)
    }
}

#[cfg(feature = "python")]
pub fn register_jsa_module(m: &Bound<'_, pyo3::types::PyModule>) -> PyResult<()> {
    m.add_function(wrap_pyfunction!(jsa_py, m)?)?;
    m.add_function(wrap_pyfunction!(jsa_batch_py, m)?)?;
    m.add_class::<JsaStreamPy>()?;
    #[cfg(feature = "cuda")]
    {
        m.add_class::<JsaDeviceArrayF32Py>()?;
        m.add_function(wrap_pyfunction!(jsa_cuda_batch_dev_py, m)?)?;
        m.add_function(wrap_pyfunction!(jsa_cuda_many_series_one_param_dev_py, m)?)?;
    }
    Ok(())
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyclass(module = "ta_indicators.cuda", unsendable)]
pub struct JsaDeviceArrayF32Py {
    buf: Option<DeviceBuffer<f32>>,
    rows: usize,
    cols: usize,
    _ctx: Arc<Context>,
    device_id: u32,
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pymethods]
impl JsaDeviceArrayF32Py {
    #[getter]
    fn __cuda_array_interface__<'py>(&self, py: Python<'py>) -> PyResult<Bound<'py, PyDict>> {
        let d = PyDict::new(py);
        d.set_item("shape", (self.rows, self.cols))?;
        d.set_item("typestr", "<f4")?;
        d.set_item(
            "strides",
            (
                self.cols * std::mem::size_of::<f32>(),
                std::mem::size_of::<f32>(),
            ),
        )?;
        let ptr = if self.rows == 0 || self.cols == 0 {
            0usize
        } else {
            self.buf
                .as_ref()
                .ok_or_else(|| PyValueError::new_err("buffer already exported via __dlpack__"))?
                .as_device_ptr()
                .as_raw() as usize
        };
        d.set_item("data", (ptr, 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<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 buf = self
            .buf
            .take()
            .ok_or_else(|| PyValueError::new_err("__dlpack__ may only be called once"))?;

        let rows = self.rows;
        let cols = self.cols;

        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"))]
impl JsaDeviceArrayF32Py {
    pub(crate) fn from_handle_rust(handle: JsaDeviceHandle) -> Self {
        JsaDeviceArrayF32Py {
            buf: Some(handle.buf),
            rows: handle.rows,
            cols: handle.cols,
            _ctx: handle._ctx,
            device_id: handle.device_id,
        }
    }
}

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

    fn check_jsa_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 = JsaParams { period: None };
        let input = JsaInput::from_candles(&candles, "close", default_params);
        let output = jsa_with_kernel(&input, kernel)?;
        assert_eq!(output.values.len(), candles.close.len());
        Ok(())
    }

    fn check_jsa_accuracy(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let expected_last_five = [61640.0, 61418.0, 61240.0, 61060.5, 60889.5];
        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;
        let default_params = JsaParams::default();
        let input = JsaInput::from_candles(&candles, "close", default_params);
        let result = jsa_with_kernel(&input, kernel)?;
        let start_idx = result.values.len() - 5;
        for (i, &val) in result.values[start_idx..].iter().enumerate() {
            let expected = expected_last_five[i];
            assert!(
                (val - expected).abs() < 1e-5,
                "[{}] mismatch idx {}: got {}, expected {}",
                test_name,
                i,
                val,
                expected
            );
        }
        Ok(())
    }

    fn check_jsa_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 = JsaParams { period: Some(0) };
        let input = JsaInput::from_slice(&input_data, params);
        let res = jsa_with_kernel(&input, kernel);
        assert!(
            res.is_err(),
            "[{}] JSA should fail with zero period",
            test_name
        );
        Ok(())
    }

    fn check_jsa_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 = JsaParams { period: Some(10) };
        let input = JsaInput::from_slice(&data_small, params);
        let res = jsa_with_kernel(&input, kernel);
        assert!(
            res.is_err(),
            "[{}] JSA should fail with period exceeding length",
            test_name
        );
        Ok(())
    }

    fn check_jsa_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 = JsaParams { period: Some(5) };
        let input = JsaInput::from_slice(&single_point, params);
        let res = jsa_with_kernel(&input, kernel);
        assert!(
            res.is_err(),
            "[{}] JSA should fail with insufficient data",
            test_name
        );
        Ok(())
    }

    fn check_jsa_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 = JsaParams { period: Some(10) };
        let first_input = JsaInput::from_candles(&candles, "close", first_params);
        let first_result = jsa_with_kernel(&first_input, kernel)?;

        let second_params = JsaParams { period: Some(5) };
        let second_input = JsaInput::from_slice(&first_result.values, second_params);
        let second_result = jsa_with_kernel(&second_input, kernel)?;

        assert_eq!(second_result.values.len(), first_result.values.len());
        for i in 30..second_result.values.len() {
            assert!(
                second_result.values[i].is_finite(),
                "[{}] NaN at idx {}",
                test_name,
                i
            );
        }
        Ok(())
    }

    fn check_jsa_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 = 30;
        let input = JsaInput::from_candles(
            &candles,
            "close",
            JsaParams {
                period: Some(period),
            },
        );
        let batch_output = jsa_with_kernel(&input, kernel)?.values;

        let mut stream = JsaStream::try_new(JsaParams {
            period: Some(period),
        })?;
        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,
                "[{}] streaming mismatch at idx {}: batch={}, stream={}, diff={}",
                test_name,
                i,
                b,
                s,
                diff
            );
        }
        Ok(())
    }

    macro_rules! generate_all_jsa_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_jsa_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![2, 5, 10, 14, 20, 30, 50, 100, 200];
        let test_sources = vec!["open", "high", "low", "close", "hl2", "hlc3", "ohlc4"];

        for period in &test_periods {
            for source in &test_sources {
                let input = JsaInput::from_candles(
                    &candles,
                    source,
                    JsaParams {
                        period: Some(*period),
                    },
                );
                let output = jsa_with_kernel(&input, kernel)?;

                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={}, source={}",
                            test_name, val, bits, i, period, source
                        );
                    }

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

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

        Ok(())
    }

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

    generate_all_jsa_tests!(
        check_jsa_partial_params,
        check_jsa_accuracy,
        check_jsa_zero_period,
        check_jsa_period_exceeds_length,
        check_jsa_very_small_dataset,
        check_jsa_reinput,
        check_jsa_streaming,
        check_jsa_no_poison
    );

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

        let def = JsaParams::default();
        let row = output.values_for(&def).expect("default row missing");
        assert_eq!(row.len(), c.close.len());

        let expected = [61640.0, 61418.0, 61240.0, 61060.5, 60889.5];
        let start = row.len() - 5;
        for (i, &v) in row[start..].iter().enumerate() {
            assert!(
                (v - expected[i]).abs() < 1e-5,
                "[{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_sources = vec!["open", "high", "low", "close", "hl2", "hlc3", "ohlc4"];

        for source in &test_sources {
            let output = JsaBatchBuilder::new()
                .kernel(kernel)
                .period_range(2, 200, 3)
                .apply_candles(&c, source)?;

            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;

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

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

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

        let edge_case_ranges = vec![(2, 5, 1), (190, 200, 2), (50, 100, 10)];
        for (start, end, step) in edge_case_ranges {
            let output = JsaBatchBuilder::new()
                .kernel(kernel)
                .period_range(start, end, 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;

                if bits == 0x11111111_11111111
                    || bits == 0x22222222_22222222
                    || bits == 0x33333333_33333333
                {
                    panic!(
						"[{}] Found poison value {} (0x{:016X}) at row {} col {} with range ({},{},{})",
						test, val, bits, row, col, start, end, step
					);
                }
            }
        }

        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);

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

        let strat = (1usize..=100).prop_flat_map(|period| {
            (
                prop::collection::vec(
                    (-1e6f64..1e6f64).prop_filter("finite", |x| x.is_finite()),
                    period..400,
                ),
                Just(period),
            )
        });

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

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

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

            let first_valid = data.iter().position(|x| !x.is_nan()).unwrap_or(0);
            let warmup_period = first_valid + period;

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

            for i in warmup_period..data.len() {
                let expected = (data[i] + data[i - period]) * 0.5;
                let actual = out[i];

                prop_assert!(
                    (actual - expected).abs() < 1e-9,
                    "[{}] Formula verification failed at index {}: expected {}, got {}, diff = {}",
                    test_name,
                    i,
                    expected,
                    actual,
                    (actual - expected).abs()
                );
            }

            for i in warmup_period..data.len() {
                let val1 = data[i];
                let val2 = data[i - period];
                let min_val = val1.min(val2);
                let max_val = val1.max(val2);
                let actual = out[i];

                prop_assert!(
                    actual >= min_val - 1e-9 && actual <= max_val + 1e-9,
                    "[{}] Output bounds check failed at index {}: {} not in [{}, {}]",
                    test_name,
                    i,
                    actual,
                    min_val,
                    max_val
                );
            }

            if kernel != Kernel::Scalar {
                for i in 0..data.len() {
                    let y = out[i];
                    let r = ref_out[i];

                    if y.is_nan() && r.is_nan() {
                        continue;
                    }

                    let y_bits = y.to_bits();
                    let r_bits = r.to_bits();
                    prop_assert!(
                        y_bits == r_bits,
                        "[{}] Cross-kernel mismatch at index {}: {} ({:016X}) != {} ({:016X})",
                        test_name,
                        i,
                        y,
                        y_bits,
                        r,
                        r_bits
                    );
                }
            }

            if data.windows(2).all(|w| (w[0] - w[1]).abs() < 1e-9) && !data.is_empty() {
                let constant = data[first_valid];
                for i in warmup_period..data.len() {
                    prop_assert!(
                        (out[i] - constant).abs() < 1e-9,
                        "[{}] Constant data test failed at index {}: expected {}, got {}",
                        test_name,
                        i,
                        constant,
                        out[i]
                    );
                }
            }

            let is_monotonic_inc = data.windows(2).all(|w| w[1] >= w[0] - 1e-12);
            if is_monotonic_inc && warmup_period + 1 < data.len() {
                for i in (warmup_period + 1)..data.len() {
                    prop_assert!(
                        out[i] >= out[i - 1] - 1e-9,
                        "[{}] Monotonic test failed at index {}: {} < {}",
                        test_name,
                        i,
                        out[i],
                        out[i - 1]
                    );
                }
            }

            if period == 1 && warmup_period < data.len() {
                for i in warmup_period..data.len() {
                    let expected = (data[i] + data[i - 1]) * 0.5;
                    let actual = out[i];
                    prop_assert!(
                        (actual - expected).abs() < 1e-9,
                        "[{}] Period=1 test failed at index {}: expected {}, got {}",
                        test_name,
                        i,
                        expected,
                        actual
                    );
                }
            }

            #[cfg(debug_assertions)]
            {
                for (i, &val) in out.iter().enumerate() {
                    if val.is_nan() {
                        continue;
                    }

                    let bits = val.to_bits();

                    prop_assert!(
                        bits != 0x11111111_11111111,
                        "[{}] Found alloc_with_nan_prefix poison at index {}",
                        test_name,
                        i
                    );
                    prop_assert!(
                        bits != 0x22222222_22222222,
                        "[{}] Found init_matrix_prefixes poison at index {}",
                        test_name,
                        i
                    );
                    prop_assert!(
                        bits != 0x33333333_33333333,
                        "[{}] Found make_uninit_matrix poison at index {}",
                        test_name,
                        i
                    );
                }
            }

            Ok(())
        })?;

        Ok(())
    }

    #[cfg(feature = "proptest")]
    generate_all_jsa_tests!(check_jsa_property);

    #[test]
    fn test_jsa_into_matches_api() -> Result<(), Box<dyn Error>> {
        let mut data = Vec::with_capacity(256);
        for _ in 0..8 {
            data.push(f64::NAN);
        }
        for i in 0..248u64 {
            let x = (i as f64).sin() * 3.14159 + (i as f64) * 0.01;
            data.push(x);
        }

        let input = JsaInput::from_slice(&data, JsaParams::default());

        let base = jsa(&input)?.values;

        let mut out = vec![0.0; data.len()];
        jsa_into(&input, &mut out)?;

        assert_eq!(base.len(), out.len(), "lengths must match");

        for (i, (&a, &b)) in base.iter().zip(out.iter()).enumerate() {
            let both_nan = a.is_nan() && b.is_nan();
            assert!(both_nan || a == b, "mismatch at idx {}: {} vs {}", i, a, b);
        }
        Ok(())
    }
}