vector_ta/indicators/moving_averages/

fwma.rs

#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::cuda_available;
#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::moving_averages::alma_wrapper::DeviceArrayF32;
#[cfg(all(feature = "python", feature = "cuda"))]
use crate::cuda::moving_averages::CudaFwma;
use crate::utilities::aligned_vector::AlignedVec;
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(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(feature = "python", feature = "cuda"))]
use cust::context::Context;
#[cfg(all(feature = "python", feature = "cuda"))]
use cust::memory::DeviceBuffer;
#[cfg(not(target_arch = "wasm32"))]
use rayon::prelude::*;
#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
use serde::{Deserialize, Serialize};
use std::convert::AsRef;
use std::error::Error;
use std::mem::{ManuallyDrop, MaybeUninit};
#[cfg(all(feature = "python", feature = "cuda"))]
use std::sync::Arc;
use thiserror::Error;
#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
use wasm_bindgen::prelude::*;

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyclass(module = "ta_indicators.cuda", name = "FwmaDeviceArrayF32", unsendable)]
pub struct DeviceArrayF32Py {
    pub(crate) inner: DeviceArrayF32,
    pub(crate) _ctx: Arc<Context>,
    pub(crate) device_id: u32,
    pub(crate) stream: usize,
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pymethods]
impl DeviceArrayF32Py {
    #[getter]
    fn __cuda_array_interface__<'py>(&self, py: Python<'py>) -> PyResult<Bound<'py, PyDict>> {
        let d = PyDict::new(py);
        d.set_item("shape", (self.inner.rows, self.inner.cols))?;
        d.set_item("typestr", "<f4")?;
        d.set_item(
            "strides",
            (
                self.inner.cols * core::mem::size_of::<f32>(),
                core::mem::size_of::<f32>(),
            ),
        )?;
        d.set_item("data", (self.inner.device_ptr() as usize, false))?;

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

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

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

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

        let dummy =
            DeviceBuffer::from_slice(&[]).map_err(|e| PyValueError::new_err(e.to_string()))?;
        let 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)
    }
}

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

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

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

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

impl Default for FwmaParams {
    fn default() -> Self {
        Self { period: Some(5) }
    }
}

#[derive(Debug, Clone)]
pub struct FwmaInput<'a> {
    pub data: FwmaData<'a>,
    pub params: FwmaParams,
}

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

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

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

impl FwmaBuilder {
    #[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<FwmaOutput, FwmaError> {
        let p = FwmaParams {
            period: self.period,
        };
        let i = FwmaInput::from_candles(c, "close", p);
        fwma_with_kernel(&i, self.kernel)
    }

    #[inline(always)]
    pub fn apply_slice(self, d: &[f64]) -> Result<FwmaOutput, FwmaError> {
        let p = FwmaParams {
            period: self.period,
        };
        let i = FwmaInput::from_slice(d, p);
        fwma_with_kernel(&i, self.kernel)
    }

    #[inline(always)]
    pub fn into_stream(self) -> Result<FwmaStream, FwmaError> {
        let p = FwmaParams {
            period: self.period,
        };
        FwmaStream::try_new(p)
    }
}

#[derive(Debug, Error)]
pub enum FwmaError {
    #[error("fwma: Input data slice is empty.")]
    EmptyInputData,
    #[error("fwma: All values are NaN.")]
    AllValuesNaN,
    #[error("fwma: Invalid period: period = {period}, data length = {data_len}")]
    InvalidPeriod { period: usize, data_len: usize },
    #[error("fwma: Not enough valid data: needed = {needed}, valid = {valid}")]
    NotEnoughValidData { needed: usize, valid: usize },
    #[error("fwma: Fibonacci sum is zero. Cannot normalize weights.")]
    ZeroFibonacciSum,
    #[error("fwma: Output buffer length mismatch: expected = {expected}, got = {got}")]
    OutputLengthMismatch { expected: usize, got: usize },
    #[error("fwma: Invalid range: start={start}, end={end}, step={step}")]
    InvalidRange {
        start: usize,
        end: usize,
        step: usize,
    },
    #[error("fwma: Invalid kernel for batch API: {0:?}")]
    InvalidKernelForBatch(Kernel),
    #[error("fwma: arithmetic overflow while computing {context}")]
    ArithmeticOverflow { context: &'static str },
}

#[inline]
pub fn fwma(input: &FwmaInput) -> Result<FwmaOutput, FwmaError> {
    fwma_with_kernel(input, Kernel::Auto)
}

#[inline(always)]
fn fwma_prepare<'a>(
    input: &'a FwmaInput,
    kernel: Kernel,
) -> Result<(&'a [f64], Vec<f64>, usize, usize, Kernel), FwmaError> {
    let data: &[f64] = input.as_ref();
    let len = data.len();
    if len == 0 {
        return Err(FwmaError::EmptyInputData);
    }
    let first = data
        .iter()
        .position(|x| !x.is_nan())
        .ok_or(FwmaError::AllValuesNaN)?;
    let period = input.get_period();

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

    let mut fib = vec![1.0; period];
    for i in 2..period {
        fib[i] = fib[i - 1] + fib[i - 2];
    }
    let fib_sum: f64 = fib.iter().sum();
    if fib_sum == 0.0 {
        return Err(FwmaError::ZeroFibonacciSum);
    }
    for w in &mut fib {
        *w /= fib_sum;
    }

    let chosen = match kernel {
        Kernel::Auto => detect_best_kernel(),
        other => other,
    };

    Ok((data, fib, period, first, chosen))
}

#[inline(always)]
fn fwma_compute_into(
    data: &[f64],
    fib: &[f64],
    period: usize,
    first: usize,
    kernel: Kernel,
    out: &mut [f64],
) {
    unsafe {
        #[cfg(all(target_arch = "wasm32", target_feature = "simd128"))]
        {
            if matches!(kernel, Kernel::Scalar | Kernel::ScalarBatch) {
                fwma_simd128(data, fib, period, first, out);
                return;
            }
        }

        match kernel {
            Kernel::Scalar | Kernel::ScalarBatch => fwma_scalar(data, fib, period, first, out),
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx2 | Kernel::Avx2Batch => fwma_avx2(data, fib, period, first, out),
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx512 | Kernel::Avx512Batch => fwma_avx512(data, fib, period, first, out),
            #[cfg(not(all(feature = "nightly-avx", target_arch = "x86_64")))]
            Kernel::Avx2 | Kernel::Avx2Batch | Kernel::Avx512 | Kernel::Avx512Batch => {
                fwma_scalar(data, fib, period, first, out)
            }
            _ => unreachable!(),
        }
    }
}

pub fn fwma_with_kernel(input: &FwmaInput, kernel: Kernel) -> Result<FwmaOutput, FwmaError> {
    let (data, fib, period, first, chosen) = fwma_prepare(input, kernel)?;

    let warm = first + period - 1;
    let mut out = alloc_with_nan_prefix(data.len(), warm);

    fwma_compute_into(data, &fib, period, first, chosen, &mut out);

    Ok(FwmaOutput { values: out })
}

#[inline]
pub fn fwma_into_slice(dst: &mut [f64], input: &FwmaInput, kern: Kernel) -> Result<(), FwmaError> {
    let (data, fib, period, first, chosen) = fwma_prepare(input, kern)?;

    if dst.len() != data.len() {
        return Err(FwmaError::OutputLengthMismatch {
            expected: data.len(),
            got: dst.len(),
        });
    }

    let warmup_end = (first + period - 1).min(dst.len());
    for v in &mut dst[..warmup_end] {
        *v = f64::from_bits(0x7ff8_0000_0000_0000);
    }

    fwma_compute_into(data, &fib, period, first, chosen, dst);

    Ok(())
}

#[cfg(not(all(target_arch = "wasm32", feature = "wasm")))]
#[inline]
pub fn fwma_into(input: &FwmaInput, out: &mut [f64]) -> Result<(), FwmaError> {
    let (data, fib, period, first, chosen) = fwma_prepare(input, Kernel::Auto)?;

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

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

    fwma_compute_into(data, &fib, period, first, chosen, out);
    Ok(())
}

#[inline(always)]
pub unsafe fn fwma_scalar(
    data: &[f64],
    fib: &[f64],
    period: usize,
    first_val: usize,
    out: &mut [f64],
) {
    assert_eq!(fib.len(), period, "fib.len() must equal period");
    assert!(
        out.len() >= data.len(),
        "out must be at least as long as data"
    );

    let p4 = period & !3;
    for i in (first_val + period - 1)..data.len() {
        let start = i + 1 - period;
        let window = &data[start..start + period];
        let mut sum = 0.0;
        for (d4, w4) in window[..p4].chunks_exact(4).zip(fib[..p4].chunks_exact(4)) {
            sum += d4[0] * w4[0] + d4[1] * w4[1] + d4[2] * w4[2] + d4[3] * w4[3];
        }
        for (d, w) in window[p4..].iter().zip(&fib[p4..]) {
            sum += d * w;
        }
        out[i] = sum;
    }
}

#[cfg(all(target_arch = "wasm32", target_feature = "simd128"))]
#[inline]
unsafe fn fwma_simd128(
    data: &[f64],
    fib: &[f64],
    period: usize,
    first_val: usize,
    out: &mut [f64],
) {
    use core::arch::wasm32::*;

    assert_eq!(fib.len(), period, "fib.len() must equal period");
    assert!(
        out.len() >= data.len(),
        "out must be at least as long as data"
    );

    const STEP: usize = 2;
    let chunks = period / STEP;
    let tail = period % STEP;

    for i in (first_val + period - 1)..data.len() {
        let start = i + 1 - period;
        let window = &data[start..start + period];

        let mut sum_vec = f64x2_splat(0.0);

        for j in 0..chunks {
            let idx = j * STEP;
            let d_vec = v128_load(&window[idx] as *const f64 as *const v128);
            let w_vec = v128_load(&fib[idx] as *const f64 as *const v128);
            sum_vec = f64x2_add(sum_vec, f64x2_mul(d_vec, w_vec));
        }

        let mut sum = f64x2_extract_lane::<0>(sum_vec) + f64x2_extract_lane::<1>(sum_vec);

        if tail > 0 {
            sum += window[chunks * STEP] * fib[chunks * STEP];
        }

        out[i] = sum;
    }
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[target_feature(enable = "avx2")]
unsafe fn horizontal_sum_avx2(v: __m256d) -> f64 {
    let high_low = _mm256_hadd_pd(v, v);
    let high = _mm256_extractf128_pd(high_low, 1);
    let low = _mm256_castpd256_pd128(high_low);
    let sum = _mm_add_pd(high, low);
    let result = _mm_hadd_pd(sum, sum);
    _mm_cvtsd_f64(result) * 0.5
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[target_feature(enable = "avx512f,fma")]
unsafe fn fwma_avx512_short(
    data: &[f64],
    fib: &[f64],
    period: usize,
    first: usize,
    out: &mut [f64],
) {
    const SIMD_WIDTH: usize = 8;
    let simd_chunks = period / SIMD_WIDTH;
    let remainder = period % SIMD_WIDTH;

    let data_ptr = data.as_ptr();
    let out_ptr = out.as_mut_ptr();

    let mut aligned_fib = AlignedVec::with_capacity(period + SIMD_WIDTH);
    let fib_buf = aligned_fib.as_mut_slice();
    fib_buf[..period].copy_from_slice(fib);
    let fib_ptr = fib_buf.as_ptr();

    let mut fib_vecs = Vec::with_capacity(simd_chunks);
    for chunk in 0..simd_chunks {
        fib_vecs.push(_mm512_load_pd(fib_ptr.add(chunk * SIMD_WIDTH)));
    }

    let tail_mask: __mmask8 = (1u8 << remainder).wrapping_sub(1);

    for idx in (first + period - 1)..data.len() {
        let start = idx + 1 - period;
        let window_ptr = data_ptr.add(start);

        _mm_prefetch(window_ptr.add(64) as *const i8, _MM_HINT_T0);

        let mut sum0 = _mm512_setzero_pd();
        let mut sum1 = _mm512_setzero_pd();
        let mut sum2 = _mm512_setzero_pd();
        let mut sum3 = _mm512_setzero_pd();

        let chunks4 = simd_chunks / 4;
        for i in 0..chunks4 {
            let base = window_ptr.add(i * 32);
            sum0 = _mm512_fmadd_pd(_mm512_loadu_pd(base), fib_vecs[i * 4 + 0], sum0);
            sum1 = _mm512_fmadd_pd(_mm512_loadu_pd(base.add(8)), fib_vecs[i * 4 + 1], sum1);
            sum2 = _mm512_fmadd_pd(_mm512_loadu_pd(base.add(16)), fib_vecs[i * 4 + 2], sum2);
            sum3 = _mm512_fmadd_pd(_mm512_loadu_pd(base.add(24)), fib_vecs[i * 4 + 3], sum3);
        }

        for i in (chunks4 * 4)..simd_chunks {
            let base = window_ptr.add(i * SIMD_WIDTH);
            sum0 = _mm512_fmadd_pd(_mm512_loadu_pd(base), fib_vecs[i], sum0);
        }

        sum0 = _mm512_add_pd(sum0, sum1);
        sum2 = _mm512_add_pd(sum2, sum3);
        sum0 = _mm512_add_pd(sum0, sum2);

        if remainder != 0 {
            let data_tail =
                _mm512_maskz_loadu_pd(tail_mask, window_ptr.add(simd_chunks * SIMD_WIDTH));
            let weight_tail =
                _mm512_maskz_load_pd(tail_mask, fib_ptr.add(simd_chunks * SIMD_WIDTH));
            sum0 = _mm512_fmadd_pd(data_tail, weight_tail, sum0);
        }

        let total = _mm512_reduce_add_pd(sum0);
        *out_ptr.add(idx) = total;
    }
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[target_feature(enable = "avx512f,fma")]
unsafe fn fwma_avx512_long(
    data: &[f64],
    fib: &[f64],
    period: usize,
    first: usize,
    out: &mut [f64],
) {
    const STEP: usize = 8;
    const UNROLL: usize = 4;

    let chunks = period / STEP;
    let tail_len = period % STEP;

    let mut aligned_fib = AlignedVec::with_capacity(period + STEP);
    let fib_buf = aligned_fib.as_mut_slice();
    fib_buf[..period].copy_from_slice(fib);
    let fib_ptr = fib_buf.as_ptr();

    let mut weight_vecs = Vec::with_capacity(chunks);
    for i in 0..chunks {
        weight_vecs.push(_mm512_load_pd(fib_ptr.add(i * STEP)));
    }

    let tmask: __mmask8 = (1u8 << tail_len).wrapping_sub(1);
    let w_tail = if tail_len > 0 {
        Some(_mm512_maskz_load_pd(tmask, fib_ptr.add(chunks * STEP)))
    } else {
        None
    };

    let end = data.len();
    let last_valid = end.saturating_sub(UNROLL - 1);
    let mut i = first + period - 1;

    while i < last_valid {
        let base0 = data.as_ptr().add(i + 1 - period);
        let base1 = base0.add(1);
        let base2 = base0.add(2);
        let base3 = base0.add(3);

        let mut sum0 = _mm512_setzero_pd();
        let mut sum1 = _mm512_setzero_pd();
        let mut sum2 = _mm512_setzero_pd();
        let mut sum3 = _mm512_setzero_pd();

        for (j, &w) in weight_vecs.iter().enumerate() {
            let offset = j * STEP;

            let d0 = _mm512_loadu_pd(base0.add(offset));
            let d1 = _mm512_loadu_pd(base1.add(offset));
            let d2 = _mm512_loadu_pd(base2.add(offset));
            let d3 = _mm512_loadu_pd(base3.add(offset));

            sum0 = _mm512_fmadd_pd(d0, w, sum0);
            sum1 = _mm512_fmadd_pd(d1, w, sum1);
            sum2 = _mm512_fmadd_pd(d2, w, sum2);
            sum3 = _mm512_fmadd_pd(d3, w, sum3);
        }

        if let Some(wt) = w_tail {
            let offset = chunks * STEP;
            let d0 = _mm512_maskz_loadu_pd(tmask, base0.add(offset));
            let d1 = _mm512_maskz_loadu_pd(tmask, base1.add(offset));
            let d2 = _mm512_maskz_loadu_pd(tmask, base2.add(offset));
            let d3 = _mm512_maskz_loadu_pd(tmask, base3.add(offset));

            sum0 = _mm512_fmadd_pd(d0, wt, sum0);
            sum1 = _mm512_fmadd_pd(d1, wt, sum1);
            sum2 = _mm512_fmadd_pd(d2, wt, sum2);
            sum3 = _mm512_fmadd_pd(d3, wt, sum3);
        }

        out[i] = _mm512_reduce_add_pd(sum0);
        out[i + 1] = _mm512_reduce_add_pd(sum1);
        out[i + 2] = _mm512_reduce_add_pd(sum2);
        out[i + 3] = _mm512_reduce_add_pd(sum3);

        i += UNROLL;
    }

    while i < end {
        let base = data.as_ptr().add(i + 1 - period);
        let mut sum = _mm512_setzero_pd();

        for (j, &w) in weight_vecs.iter().enumerate() {
            let d = _mm512_loadu_pd(base.add(j * STEP));
            sum = _mm512_fmadd_pd(d, w, sum);
        }

        if let Some(wt) = w_tail {
            let d = _mm512_maskz_loadu_pd(tmask, base.add(chunks * STEP));
            sum = _mm512_fmadd_pd(d, wt, sum);
        }

        out[i] = _mm512_reduce_add_pd(sum);
        i += 1;
    }
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[target_feature(enable = "avx512f,fma")]
pub unsafe fn fwma_avx512(data: &[f64], fib: &[f64], period: usize, first: usize, out: &mut [f64]) {
    if period <= 32 {
        fwma_avx512_short(data, fib, period, first, out);
    } else {
        fwma_avx512_long(data, fib, period, first, out);
    }
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[target_feature(enable = "avx2")]
pub unsafe fn fwma_avx2(
    data: &[f64],
    fib: &[f64],
    period: usize,
    first_valid: usize,
    out: &mut [f64],
) {
    const W: usize = 4;
    let full = period / W;
    let tail = period % W;

    let mut aligned = AlignedVec::with_capacity(period + W);
    let fib_aln = aligned.as_mut_slice();
    fib_aln[..period].copy_from_slice(fib);
    let wptr = fib_aln.as_ptr();

    let dptr = data.as_ptr();
    let optr = out.as_mut_ptr();

    for i in (first_valid + period - 1)..data.len() {
        let base = dptr.add(i + 1 - period);

        let mut acc0 = _mm256_setzero_pd();
        let mut acc1 = _mm256_setzero_pd();

        let mut j = 0;
        while j + 8 <= period {
            let v0 = _mm256_loadu_pd(base.add(j));
            let w0 = _mm256_load_pd(wptr.add(j));
            acc0 = _mm256_fmadd_pd(v0, w0, acc0);

            let v1 = _mm256_loadu_pd(base.add(j + 4));
            let w1 = _mm256_load_pd(wptr.add(j + 4));
            acc1 = _mm256_fmadd_pd(v1, w1, acc1);

            j += 8;
        }
        if j + 4 <= period {
            let v = _mm256_loadu_pd(base.add(j));
            let w = _mm256_load_pd(wptr.add(j));
            acc0 = _mm256_fmadd_pd(v, w, acc0);
            j += 4;
        }

        let sum_vec = _mm256_add_pd(acc0, acc1);
        let mut sum = horizontal_sum_avx2(sum_vec);

        for k in 0..tail {
            let idx = period - tail + k;
            sum += *base.add(idx) * *wptr.add(idx);
        }
        *optr.add(i) = sum;
    }
}

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

    w: Vec<f64>,
    w0: f64,
    w_last: f64,
    w_prev: f64,
    w_next: f64,

    buffer: Vec<f64>,
    head: usize,
    filled: bool,

    acc_n: f64,
    acc_d: f64,

    nan_count: usize,
}

impl FwmaStream {
    pub fn try_new(params: FwmaParams) -> Result<Self, FwmaError> {
        let period = params.period.unwrap_or(5);
        if period == 0 {
            return Err(FwmaError::InvalidPeriod {
                period,
                data_len: 0,
            });
        }

        let mut w = vec![1.0; period];
        for i in 2..period {
            w[i] = w[i - 1] + w[i - 2];
        }

        let sum: f64 = w.iter().sum();
        if sum == 0.0 {
            return Err(FwmaError::ZeroFibonacciSum);
        }
        for wi in &mut w {
            *wi /= sum;
        }

        let w0 = w[0];
        let w_last = w[period - 1];
        let w_prev = if period > 1 { w[period - 2] } else { 0.0 };
        let w_next = w_last + w_prev;

        Ok(Self {
            period,
            w,
            w0,
            w_last,
            w_prev,
            w_next,
            buffer: vec![f64::NAN; period],
            head: 0,
            filled: false,
            acc_n: 0.0,
            acc_d: 0.0,
            nan_count: 0,
        })
    }

    #[inline(always)]
    pub fn update(&mut self, value: f64) -> Option<f64> {
        let old_raw = self.buffer[self.head];

        if self.filled && old_raw.is_nan() {
            self.nan_count = self.nan_count.saturating_sub(1);
        }

        self.buffer[self.head] = value;
        if value.is_nan() {
            self.nan_count += 1;
        }

        self.head += 1;
        if self.head == self.period {
            self.head = 0;
        }

        if !self.filled {
            if self.head != 0 {
                return None;
            }

            self.filled = true;

            let mut n = 0.0f64;
            let mut q = 0.0f64;
            let last = self.period - 1;
            for j in 0..self.period {
                let x = self.buffer[j];
                let xv = if x.is_nan() { 0.0 } else { x };

                n += xv * self.w[j];

                let wn = if j < last { self.w[j + 1] } else { self.w_next };
                q += xv * wn;
            }
            self.acc_n = n;
            self.acc_d = q - n;

            return Some(if self.nan_count == 0 { n } else { f64::NAN });
        }

        let x_old = if old_raw.is_nan() { 0.0 } else { old_raw };
        let x_new = if value.is_nan() { 0.0 } else { value };

        let prev_n = self.acc_n;

        let d_old = self.acc_d;
        let n_prime = x_new.mul_add(self.w_last, d_old);

        let d_prime = x_new.mul_add(self.w_prev, prev_n - d_old - self.w0 * x_old);

        self.acc_n = n_prime;
        self.acc_d = d_prime;

        Some(if self.nan_count == 0 {
            self.dot_ring()
        } else {
            f64::NAN
        })
    }
}

impl FwmaStream {
    #[inline(always)]
    fn dot_ring(&self) -> f64 {
        let mut sum = 0.0;
        let mut idx = self.head;
        for &wj in &self.w {
            sum += wj * self.buffer[idx];
            idx += 1;
            if idx == self.period {
                idx = 0;
            }
        }
        sum
    }
}

#[derive(Clone, Debug)]
pub struct FwmaBatchRange {
    pub period: (usize, usize, usize),
}
impl Default for FwmaBatchRange {
    fn default() -> Self {
        Self {
            period: (5, 254, 1),
        }
    }
}

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

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

pub fn fwma_batch_with_kernel(
    data: &[f64],
    sweep: &FwmaBatchRange,
    k: Kernel,
) -> Result<FwmaBatchOutput, FwmaError> {
    let kernel = match k {
        Kernel::Auto => detect_best_batch_kernel(),
        other if other.is_batch() => other,
        other => {
            return Err(FwmaError::InvalidKernelForBatch(other));
        }
    };
    let simd = match kernel {
        #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
        Kernel::Avx512Batch => Kernel::Avx512,
        #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
        Kernel::Avx2Batch => Kernel::Avx2,
        Kernel::ScalarBatch => Kernel::Scalar,
        _ => unreachable!(),
    };
    fwma_batch_par_slice(data, sweep, simd)
}

#[derive(Clone, Debug)]
pub struct FwmaBatchOutput {
    pub values: Vec<f64>,
    pub combos: Vec<FwmaParams>,
    pub rows: usize,
    pub cols: usize,
}
impl FwmaBatchOutput {
    pub fn row_for_params(&self, p: &FwmaParams) -> Option<usize> {
        self.combos
            .iter()
            .position(|c| c.period.unwrap_or(5) == p.period.unwrap_or(5))
    }
    pub fn values_for(&self, p: &FwmaParams) -> 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: &FwmaBatchRange) -> Vec<FwmaParams> {
    fn axis_usize((start, end, step): (usize, usize, usize)) -> Vec<usize> {
        if step == 0 || start == end {
            return vec![start];
        }
        let (lo, hi) = if start <= end {
            (start, end)
        } else {
            (end, start)
        };
        (lo..=hi).step_by(step).collect()
    }
    let periods = axis_usize(r.period);
    let mut out = Vec::with_capacity(periods.len());
    for &p in &periods {
        out.push(FwmaParams { period: Some(p) });
    }
    out
}

#[inline(always)]
fn fill_nan_prefixes_slice(
    rows: usize,
    cols: usize,
    warmup_periods: &[usize],
    out_slice: &mut [f64],
) {
    for (row, &warmup) in warmup_periods.iter().enumerate() {
        let row_start = row * cols;
        let row_end = row_start + warmup.min(cols);
        for i in row_start..row_end {
            out_slice[i] = f64::NAN;
        }
    }
}

#[inline(always)]
pub fn fwma_batch_slice(
    data: &[f64],
    sweep: &FwmaBatchRange,
    kern: Kernel,
) -> Result<FwmaBatchOutput, FwmaError> {
    fwma_batch_inner(data, sweep, kern, false)
}

#[inline(always)]
pub fn fwma_batch_par_slice(
    data: &[f64],
    sweep: &FwmaBatchRange,
    kern: Kernel,
) -> Result<FwmaBatchOutput, FwmaError> {
    fwma_batch_inner(data, sweep, kern, true)
}

#[inline(always)]
fn fwma_batch_inner(
    data: &[f64],
    sweep: &FwmaBatchRange,
    kern: Kernel,
    parallel: bool,
) -> Result<FwmaBatchOutput, FwmaError> {
    let combos = expand_grid(sweep);
    if combos.is_empty() {
        let (s, e, t) = sweep.period;
        return Err(FwmaError::InvalidRange {
            start: s,
            end: e,
            step: t,
        });
    }

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

    let _total = rows
        .checked_mul(cols)
        .ok_or(FwmaError::ArithmeticOverflow {
            context: "rows*cols in fwma_batch_inner",
        })?;

    let mut buf_mu = make_uninit_matrix(rows, cols);

    let first = data.iter().position(|x| !x.is_nan()).unwrap_or(0);
    let warm: Vec<usize> = combos
        .iter()
        .map(|c| first + c.period.unwrap() - 1)
        .collect();

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

    let mut buf_guard = ManuallyDrop::new(buf_mu);
    let values_slice: &mut [f64] = unsafe {
        core::slice::from_raw_parts_mut(buf_guard.as_mut_ptr() as *mut f64, buf_guard.len())
    };

    fwma_batch_inner_into(data, sweep, kern, parallel, values_slice)?;

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

    Ok(FwmaBatchOutput {
        values,
        combos,
        rows,
        cols,
    })
}
#[inline(always)]
fn fwma_batch_inner_into(
    data: &[f64],
    sweep: &FwmaBatchRange,
    kern: Kernel,
    parallel: bool,
    out: &mut [f64],
) -> Result<Vec<FwmaParams>, FwmaError> {
    let combos = expand_grid(sweep);
    if combos.is_empty() {
        let (s, e, t) = sweep.period;
        return Err(FwmaError::InvalidRange {
            start: s,
            end: e,
            step: t,
        });
    }

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

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

    let cap = rows
        .checked_mul(max_p)
        .ok_or(FwmaError::ArithmeticOverflow {
            context: "rows*max_p in fwma_batch_inner_into",
        })?;
    let mut aligned = AlignedVec::with_capacity(cap);
    let flat_fib = aligned.as_mut_slice();

    for (row, prm) in combos.iter().enumerate() {
        let period = prm.period.unwrap();
        let base = row * max_p;
        let slice = &mut flat_fib[base..base + period];
        slice[0] = 1.0;
        if period > 1 {
            slice[1] = 1.0;
        }
        for i in 2..period {
            slice[i] = slice[i - 1] + slice[i - 2];
        }
        let sum: f64 = slice[..period].iter().sum();
        for w in &mut slice[..period] {
            *w /= sum;
        }
    }

    let do_row = |row: usize, dst: &mut [f64]| unsafe {
        let period = combos[row].period.unwrap();
        let fib_ptr = flat_fib.as_ptr().add(row * max_p);

        match kern {
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx512 => fwma_row_avx512(data, first, period, max_p, fib_ptr, dst),
            #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
            Kernel::Avx2 => fwma_row_avx2(data, first, period, max_p, fib_ptr, dst),
            _ => fwma_row_scalar(data, first, period, max_p, fib_ptr, dst),
        }
    };

    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)
}

#[inline]
unsafe fn fwma_row_scalar(
    data: &[f64],
    first: usize,
    period: usize,
    _stride: usize,
    fib_ptr: *const f64,
    out: &mut [f64],
) {
    let fib = std::slice::from_raw_parts(fib_ptr, period);
    fwma_scalar(data, fib, period, first, out);
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[target_feature(enable = "avx2")]
#[inline]
unsafe fn fwma_row_avx2(
    data: &[f64],
    first: usize,
    period: usize,
    _stride: usize,
    fib_ptr: *const f64,
    out: &mut [f64],
) {
    let fib = std::slice::from_raw_parts(fib_ptr, period);
    fwma_avx2(data, fib, period, first, out);
}

#[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
#[target_feature(enable = "avx512f,fma")]
#[inline]
unsafe fn fwma_row_avx512(
    data: &[f64],
    first: usize,
    period: usize,
    _stride: usize,
    fib_ptr: *const f64,
    out: &mut [f64],
) {
    let fib = std::slice::from_raw_parts(fib_ptr, period);
    fwma_avx512(data, fib, period, first, out);
}

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

    fn check_fwma_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 = FwmaParams { period: None };
        let input = FwmaInput::from_candles(&candles, "close", default_params);
        let output = fwma_with_kernel(&input, kernel)?;
        assert_eq!(output.values.len(), candles.close.len());

        Ok(())
    }

    fn check_fwma_accuracy(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;

        let input = FwmaInput::with_default_candles(&candles);
        let result = fwma_with_kernel(&input, kernel)?;
        let expected_last_five = [
            59273.583333333336,
            59252.5,
            59167.083333333336,
            59151.0,
            58940.333333333336,
        ];
        let start = result.values.len().saturating_sub(5);
        for (i, &val) in result.values[start..].iter().enumerate() {
            let diff = (val - expected_last_five[i]).abs();
            assert!(
                diff < 1e-8,
                "[{}] FWMA {:?} mismatch at idx {}: got {}, expected {}",
                test_name,
                kernel,
                i,
                val,
                expected_last_five[i]
            );
        }
        Ok(())
    }

    fn check_fwma_default_candles(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;

        let input = FwmaInput::with_default_candles(&candles);
        match input.data {
            FwmaData::Candles { source, .. } => assert_eq!(source, "close"),
            _ => panic!("Expected FwmaData::Candles"),
        }
        let output = fwma_with_kernel(&input, kernel)?;
        assert_eq!(output.values.len(), candles.close.len());

        Ok(())
    }

    fn check_fwma_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 = FwmaParams { period: Some(0) };
        let input = FwmaInput::from_slice(&input_data, params);
        let res = fwma_with_kernel(&input, kernel);
        assert!(
            res.is_err(),
            "[{}] FWMA should fail with zero period",
            test_name
        );
        Ok(())
    }

    fn check_fwma_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 = FwmaParams { period: Some(10) };
        let input = FwmaInput::from_slice(&data_small, params);
        let res = fwma_with_kernel(&input, kernel);
        assert!(
            res.is_err(),
            "[{}] FWMA should fail with period exceeding length",
            test_name
        );
        Ok(())
    }

    fn check_fwma_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 = FwmaParams { period: Some(5) };
        let input = FwmaInput::from_slice(&single_point, params);
        let res = fwma_with_kernel(&input, kernel);
        assert!(
            res.is_err(),
            "[{}] FWMA should fail with insufficient data",
            test_name
        );
        Ok(())
    }

    fn check_fwma_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 = FwmaParams { period: Some(5) };
        let first_input = FwmaInput::from_candles(&candles, "close", first_params);
        let first_result = fwma_with_kernel(&first_input, kernel)?;

        let second_params = FwmaParams { period: Some(3) };
        let second_input = FwmaInput::from_slice(&first_result.values, second_params);
        let second_result = fwma_with_kernel(&second_input, kernel)?;

        assert_eq!(second_result.values.len(), first_result.values.len());
        for i in 240..second_result.values.len() {
            assert!(
                !second_result.values[i].is_nan(),
                "[{}] NaN found at idx {}",
                test_name,
                i
            );
        }
        Ok(())
    }

    fn check_fwma_nan_handling(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);
        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;

        let input = FwmaInput::from_candles(&candles, "close", FwmaParams { period: Some(5) });
        let res = fwma_with_kernel(&input, kernel)?;
        assert_eq!(res.values.len(), candles.close.len());
        if res.values.len() > 50 {
            for (i, &val) in res.values[50..].iter().enumerate() {
                assert!(
                    !val.is_nan(),
                    "[{}] Found unexpected NaN at out-index {}",
                    test_name,
                    50 + i
                );
            }
        }
        Ok(())
    }

    fn check_fwma_streaming(test_name: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test_name);

        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path)?;

        let period = 5;

        let input = FwmaInput::from_candles(
            &candles,
            "close",
            FwmaParams {
                period: Some(period),
            },
        );
        let batch_output = fwma_with_kernel(&input, kernel)?.values;

        let mut stream = FwmaStream::try_new(FwmaParams {
            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,
                "[{}] FWMA streaming f64 mismatch at idx {}: batch={}, stream={}, diff={}",
                test_name,
                i,
                b,
                s,
                diff
            );
        }
        Ok(())
    }

    #[cfg(debug_assertions)]
    fn check_fwma_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_cases = vec![
            FwmaParams::default(),
            FwmaParams { period: Some(2) },
            FwmaParams { period: Some(3) },
            FwmaParams { period: Some(5) },
            FwmaParams { period: Some(8) },
            FwmaParams { period: Some(10) },
            FwmaParams { period: Some(15) },
            FwmaParams { period: Some(20) },
            FwmaParams { period: Some(30) },
            FwmaParams { period: Some(50) },
        ];

        for params in test_cases {
            let input = FwmaInput::from_candles(&candles, "close", params.clone());
            let output = fwma_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 params period={:?}",
                        test_name, val, bits, i, params.period
                    );
                }

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

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

        Ok(())
    }

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

    macro_rules! generate_all_fwma_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);
                    }

                    #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
                    #[test]
                    fn [<$test_fn _avx512_f64>]() {
                        let _ = $test_fn(stringify!([<$test_fn _avx512_f64>]), Kernel::Avx512);
                    }


                    #[cfg(all(target_arch = "wasm32", target_feature = "simd128"))]
                    #[test]
                    fn [<$test_fn _simd128_f64>]() {
                        let _ = $test_fn(stringify!([<$test_fn _simd128_f64>]), Kernel::Scalar);
                    }
                )*
            }
        }
    }

    generate_all_fwma_tests!(
        check_fwma_partial_params,
        check_fwma_accuracy,
        check_fwma_default_candles,
        check_fwma_zero_period,
        check_fwma_period_exceeds_length,
        check_fwma_very_small_dataset,
        check_fwma_reinput,
        check_fwma_nan_handling,
        check_fwma_streaming,
        check_fwma_no_poison
    );

    #[cfg(feature = "proptest")]
    generate_all_fwma_tests!(check_fwma_property);

    #[test]
    fn test_fwma_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 input = FwmaInput::with_default_candles(&candles);
        let baseline = fwma(&input)?.values;

        let mut out = vec![0.0f64; baseline.len()];

        #[cfg(not(all(target_arch = "wasm32", feature = "wasm")))]
        {
            fwma_into(&input, &mut out)?;
        }
        #[cfg(all(target_arch = "wasm32", feature = "wasm"))]
        {
            fwma_into_slice(&mut out, &input, Kernel::Auto)?;
        }

        assert_eq!(out.len(), baseline.len());

        for (i, (&a, &b)) in out.iter().zip(baseline.iter()).enumerate() {
            let equal = (a.is_nan() && b.is_nan()) || (a == b);
            assert!(
                equal,
                "into parity mismatch at idx {}: got {}, expected {}",
                i, a, b
            );
        }

        Ok(())
    }

    fn check_batch_default_row(test: &str, kernel: Kernel) -> Result<(), Box<dyn Error>> {
        skip_if_unsupported!(kernel, test);

        let file = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let c = read_candles_from_csv(file)?;

        let output = FwmaBatchBuilder::new()
            .kernel(kernel)
            .apply_candles(&c, "close")?;

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

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

        let expected = [
            59273.583333333336,
            59252.5,
            59167.083333333336,
            59151.0,
            58940.333333333336,
        ];
        let start = row.len() - 5;
        for (i, &v) in row[start..].iter().enumerate() {
            assert!(
                (v - expected[i]).abs() < 1e-8,
                "[{test}] default-row mismatch at idx {i}: {v} vs {expected:?}"
            );
        }
        Ok(())
    }

    #[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, 5, 1),
            (5, 15, 2),
            (10, 30, 5),
            (3, 10, 1),
            (5, 50, 10),
            (2, 20, 1),
        ];

        for (start, end, step) in test_configs {
            let output = FwmaBatchBuilder::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;
                let params = &output.combos[row];

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

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

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

        Ok(())
    }

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

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

        let strat = (1usize..=64).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, |(mut data, period)| {
                if data.len() > period && period > 1 {
                    if data.len() % 10 == 0 {
                        data.truncate(period);
                    }
                }

                let params = FwmaParams {
                    period: Some(period),
                };
                let input = FwmaInput::from_slice(&data, params);

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

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

                prop_assert_eq!(out.len(), data.len());
                prop_assert_eq!(ref_out.len(), data.len());

                for i in 0..(period - 1).min(data.len()) {
                    prop_assert!(
                        out[i].is_nan(),
                        "Expected NaN during warmup at index {}, got {}",
                        i,
                        out[i]
                    );
                }

                if period == 2 && data.len() >= 2 {
                    let expected = (data[0] + data[1]) / 2.0;
                    if out[1].is_finite() && data[0].is_finite() && data[1].is_finite() {
                        prop_assert!(
                            (out[1] - expected).abs() <= 1e-9,
                            "Period=2: output {} should equal average {} at index 1",
                            out[1],
                            expected
                        );
                    }
                }

                for i in (period - 1)..data.len() {
                    let window = &data[i + 1 - period..=i];
                    let lo = window.iter().cloned().fold(f64::INFINITY, f64::min);
                    let hi = window.iter().cloned().fold(f64::NEG_INFINITY, f64::max);
                    let y = out[i];
                    let r = ref_out[i];

                    prop_assert!(
                        y.is_nan() || (y >= lo - 1e-9 && y <= hi + 1e-9),
                        "idx {}: {} ∉ [{}, {}]",
                        i,
                        y,
                        lo,
                        hi
                    );

                    if period == 1 {
                        prop_assert!(
                            (y - data[i]).abs() <= f64::EPSILON,
                            "Period=1: output {} should equal input {} at index {}",
                            y,
                            data[i],
                            i
                        );
                    }

                    if data.windows(2).all(|w| w[0] == w[1]) && !data.is_empty() {
                        prop_assert!(
                            (y - data[0]).abs() <= 1e-9,
                            "Constant data: output {} should equal constant {} at index {}",
                            y,
                            data[0],
                            i
                        );
                    }

                    if window.iter().any(|x| x.is_nan()) {
                        prop_assert!(
                            y.is_nan(),
                            "Window contains NaN but output {} is not NaN at index {}",
                            y,
                            i
                        );
                    }

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

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

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

                let is_monotonic_inc = data.windows(2).all(|w| w[0] <= w[1]);
                let is_monotonic_dec = data.windows(2).all(|w| w[0] >= w[1]);

                if (is_monotonic_inc || is_monotonic_dec) && data.len() >= period + 1 {
                    for i in period..out.len() {
                        if out[i].is_finite() && out[i - 1].is_finite() {
                            if is_monotonic_inc {
                                prop_assert!(
									out[i] >= out[i-1] - 1e-9,
									"Monotonic increasing data but output decreases: {} < {} at index {}",
									out[i],
									out[i-1],
									i
								);
                            }
                            if is_monotonic_dec {
                                prop_assert!(
									out[i] <= out[i-1] + 1e-9,
									"Monotonic decreasing data but output increases: {} > {} at index {}",
									out[i],
									out[i-1],
									i
								);
                            }
                        }
                    }
                }

                if period >= 3 && data.len() >= period * 2 {
                    let test_start = period;
                    if test_start + period <= data.len() {
                        let all_ascending = (0..period).all(|j| {
                            let idx = test_start + j;
                            idx == 0
                                || !data[idx].is_finite()
                                || !data[idx - 1].is_finite()
                                || data[idx] >= data[idx - 1]
                        });

                        if all_ascending && out[test_start + period - 1].is_finite() {
                            let window = &data[test_start..test_start + period];
                            let window_avg = window.iter().sum::<f64>() / period as f64;
                            if window.iter().all(|x| x.is_finite()) {
                                prop_assert!(
                                    out[test_start + period - 1] >= window_avg - 1e-9,
                                    "FWMA {} should be >= average {} for ascending window",
                                    out[test_start + period - 1],
                                    window_avg
                                );
                            }
                        }
                    }
                }

                if period > 1 && data.len() >= period * 2 {
                    let data_range = data.iter().fold(f64::NEG_INFINITY, |a, &b| a.max(b.abs()));
                    for &val in &out[(period - 1)..] {
                        if val.is_finite() && data_range > 0.0 {
                            prop_assert!(
                                val.abs() <= data_range * 1.1,
                                "Output {} exceeds reasonable bounds for data range {}",
                                val,
                                data_range
                            );
                        }
                    }
                }

                if period == 3 && data.len() >= 3 {
                    let idx = period - 1;
                    if data[idx - 2].is_finite()
                        && data[idx - 1].is_finite()
                        && data[idx].is_finite()
                    {
                        let expected =
                            data[idx - 2] * 0.25 + data[idx - 1] * 0.25 + data[idx] * 0.5;
                        prop_assert!(
                            (out[idx] - expected).abs() <= 1e-9,
                            "Period=3: output {} should equal weighted avg {} at index {}",
                            out[idx],
                            expected,
                            idx
                        );
                    }
                }

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

        let nan_strat = (2usize..=10).prop_flat_map(|period| (Just(period), 1usize..10));

        proptest::test_runner::TestRunner::default()
            .run(&nan_strat, |(period, nan_pos)| {
                let mut data = vec![
                    1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0,
                ];
                if nan_pos < data.len() {
                    data[nan_pos] = f64::NAN;
                }

                let params = FwmaParams {
                    period: Some(period),
                };
                let input = FwmaInput::from_slice(&data, params);
                let FwmaOutput { values: out } = fwma_with_kernel(&input, kernel).unwrap();

                for i in (period - 1)..data.len() {
                    let window_start = i + 1 - period;
                    let window = &data[window_start..=i];
                    let has_nan = window.iter().any(|x| x.is_nan());

                    if has_nan {
                        prop_assert!(
                            out[i].is_nan(),
                            "Window [{}, {}] contains NaN but output {} is not NaN at index {}",
                            window_start,
                            i,
                            out[i],
                            i
                        );
                    }
                }
                Ok(())
            })
            .unwrap();

        let extreme_strat = (1usize..=10).prop_flat_map(|period| (Just(period), prop::bool::ANY));

        proptest::test_runner::TestRunner::default()
            .run(&extreme_strat, |(period, use_max)| {
                let extreme_val = if use_max { 1e308 } else { 1e-308 };
                let data = vec![extreme_val; period * 2];

                let params = FwmaParams {
                    period: Some(period),
                };
                let input = FwmaInput::from_slice(&data, params);
                let result = fwma_with_kernel(&input, kernel);

                prop_assert!(result.is_ok(), "Failed to handle extreme values");

                if let Ok(FwmaOutput { values: out }) = result {
                    for i in (period - 1)..data.len() {
                        if out[i].is_finite() {
                            prop_assert!(
                                (out[i] - extreme_val).abs() / extreme_val.abs() <= 1e-9,
                                "Extreme constant value {} doesn't match output {} at index {}",
                                extreme_val,
                                out[i],
                                i
                            );
                        }
                    }
                }
                Ok(())
            })
            .unwrap();

        Ok(())
    }
    macro_rules! gen_batch_tests {
        ($fn_name:ident) => {
            paste::paste! {
                #[test]
                fn [<$fn_name _scalar>]() {
                    let _ = $fn_name(stringify!([<$fn_name _scalar>]), Kernel::ScalarBatch);
                }
                #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
                #[test]
                fn [<$fn_name _avx2>]() {
                    let _ = $fn_name(stringify!([<$fn_name _avx2>]), Kernel::Avx2Batch);
                }
                #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
                #[test]
                fn [<$fn_name _avx512>]() {
                    let _ = $fn_name(stringify!([<$fn_name _avx512>]), Kernel::Avx512Batch);
                }
                #[test]
                fn [<$fn_name _auto_detect>]() {
                    let _ = $fn_name(stringify!([<$fn_name _auto_detect>]), Kernel::Auto);
                }
            }
        };
    }
    gen_batch_tests!(check_batch_default_row);
    gen_batch_tests!(check_batch_no_poison);

    #[cfg(all(target_arch = "wasm32", feature = "wasm"))]
    #[test]
    fn test_fwma_batch_into_warmup() {
        let data = vec![1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0];
        let len = data.len();

        let period_start = 2;
        let period_end = 4;
        let period_step = 1;

        let sweep = FwmaBatchRange {
            period: (period_start, period_end, period_step),
        };
        let combos = expand_grid(&sweep);
        let rows = combos.len();

        let mut output = vec![999.0; rows * len];

        let result = unsafe {
            fwma_batch_into(
                data.as_ptr(),
                output.as_mut_ptr(),
                len,
                period_start,
                period_end,
                period_step,
            )
        };

        assert!(result.is_ok());
        assert_eq!(result.unwrap(), rows);

        for (r, params) in combos.iter().enumerate() {
            let period = params.period.unwrap();
            let warmup_end = period - 1;

            for i in 0..warmup_end {
                let idx = r * len + i;
                assert!(
                    output[idx].is_nan(),
                    "Expected NaN at row {} col {} (period {}) but got {}",
                    r,
                    i,
                    period,
                    output[idx]
                );
            }

            let first_valid_idx = r * len + warmup_end;
            assert!(
                !output[first_valid_idx].is_nan(),
                "Expected valid value at row {} col {} (period {}) but got NaN",
                r,
                warmup_end,
                period
            );
        }
    }

    #[test]
    fn print_actual_fwma_values() {
        let file_path = "src/data/2018-09-01-2024-Bitfinex_Spot-4h.csv";
        let candles = read_candles_from_csv(file_path).unwrap();

        println!("\nCandle data info:");
        println!("Total candles: {}", candles.close.len());
        println!("Last 10 close prices:");
        let close_len = candles.close.len();
        for i in (close_len - 10)..close_len {
            println!("  [{}]: {}", i - (close_len - 10), candles.close[i]);
        }

        let input = FwmaInput::with_default_candles(&candles);
        let result = fwma(&input).unwrap();

        println!("\nFWMA results (period = 5):");
        println!("Last 5 values:");
        let result_len = result.values.len();
        for i in (result_len - 5)..result_len {
            println!("  [{}]: {:.12}", i - (result_len - 5), result.values[i]);
        }

        let expected_last_five = [
            59273.583333333336,
            59252.5,
            59167.083333333336,
            59151.0,
            58940.333333333336,
        ];

        println!("\nExpected values:");
        for (i, val) in expected_last_five.iter().enumerate() {
            println!("  [{}]: {:.12}", i, val);
        }

        println!("\nDifferences:");
        for i in 0..5 {
            let actual = result.values[result_len - 5 + i];
            let expected = expected_last_five[i];
            println!(
                "  [{}]: {:.12} (diff: {:.2e})",
                i,
                actual - expected,
                (actual - expected).abs()
            );
        }
    }
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyfunction(name = "fwma_cuda_batch_dev")]
#[pyo3(signature = (data, period_range, device_id=0))]
pub fn fwma_cuda_batch_dev_py(
    py: Python<'_>,
    data: PyReadonlyArray1<'_, f64>,
    period_range: (usize, usize, usize),
    device_id: usize,
) -> PyResult<DeviceArrayF32Py> {
    use numpy::PyArrayMethods;

    if !cuda_available() {
        return Err(PyValueError::new_err("CUDA not available"));
    }

    let slice_in = data.as_slice()?;
    let sweep = FwmaBatchRange {
        period: period_range,
    };
    let data_f32: Vec<f32> = slice_in.iter().map(|&v| v as f32).collect();

    let (inner, ctx, dev_id) = py.allow_threads(|| {
        let cuda = CudaFwma::new(device_id).map_err(|e| PyValueError::new_err(e.to_string()))?;
        let ctx = cuda.context_arc_clone();
        let dev = cuda.device_id();
        let arr = cuda
            .fwma_batch_dev(&data_f32, &sweep)
            .map_err(|e| PyValueError::new_err(e.to_string()))?;
        Ok::<_, PyErr>((arr, ctx, dev))
    })?;

    Ok(DeviceArrayF32Py {
        inner,
        _ctx: ctx,
        device_id: dev_id,
        stream: 0,
    })
}

#[cfg(all(feature = "python", feature = "cuda"))]
#[pyfunction(name = "fwma_cuda_many_series_one_param_dev")]
#[pyo3(signature = (data_tm_f32, period, device_id=0))]
pub fn fwma_cuda_many_series_one_param_dev_py(
    py: Python<'_>,
    data_tm_f32: numpy::PyReadonlyArray2<'_, f32>,
    period: usize,
    device_id: usize,
) -> PyResult<DeviceArrayF32Py> {
    use numpy::PyUntypedArrayMethods;

    if !cuda_available() {
        return Err(PyValueError::new_err("CUDA not available"));
    }

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

    let (inner, ctx, dev_id) = py.allow_threads(|| {
        let cuda = CudaFwma::new(device_id).map_err(|e| PyValueError::new_err(e.to_string()))?;
        let ctx = cuda.context_arc_clone();
        let dev = cuda.device_id();
        let arr = cuda
            .fwma_multi_series_one_param_time_major_dev(flat_in, cols, rows, &params)
            .map_err(|e| PyValueError::new_err(e.to_string()))?;
        Ok::<_, PyErr>((arr, ctx, dev))
    })?;

    Ok(DeviceArrayF32Py {
        inner,
        _ctx: ctx,
        device_id: dev_id,
        stream: 0,
    })
}

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

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

    let params = FwmaParams {
        period: Some(period),
    };
    let fwma_in = FwmaInput::from_slice(slice_in, params);

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

    Ok(result_vec.into_pyarray(py))
}

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

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

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

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

    let slice_in = data.as_slice()?;

    let sweep = FwmaBatchRange {
        period: period_range,
    };

    let combos = expand_grid(&sweep);
    let rows = combos.len();
    let cols = slice_in.len();

    let total = rows
        .checked_mul(cols)
        .ok_or_else(|| PyValueError::new_err("fwma_batch: rows*cols overflow"))?;
    let out_arr = unsafe { PyArray1::<f64>::new(py, [total], false) };
    let slice_out = unsafe { out_arr.as_slice_mut()? };

    let kern = validate_kernel(kernel, true)?;

    let first = slice_in.iter().position(|x| !x.is_nan()).unwrap_or(0);
    let warm: Vec<usize> = combos
        .iter()
        .map(|c| first + c.period.unwrap() - 1)
        .collect();

    fill_nan_prefixes_slice(rows, cols, &warm, slice_out);

    let combos = py
        .allow_threads(|| {
            let kernel = match kern {
                Kernel::Auto => detect_best_batch_kernel(),
                k => k,
            };
            let simd = match kernel {
                #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
                Kernel::Avx512Batch => Kernel::Avx512,
                #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
                Kernel::Avx2Batch => Kernel::Avx2,
                Kernel::ScalarBatch => Kernel::Scalar,
                #[cfg(not(all(feature = "nightly-avx", target_arch = "x86_64")))]
                _ => Kernel::Scalar,
                #[cfg(all(feature = "nightly-avx", target_arch = "x86_64"))]
                _ => unreachable!(),
            };
            fwma_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(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn fwma_js(data: &[f64], period: usize) -> Result<Vec<f64>, JsValue> {
    let params = FwmaParams {
        period: Some(period),
    };
    let input = FwmaInput::from_slice(data, params);

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

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

    Ok(output)
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn fwma_batch_js(
    data: &[f64],
    period_start: usize,
    period_end: usize,
    period_step: usize,
) -> Result<Vec<f64>, JsValue> {
    let sweep = FwmaBatchRange {
        period: (period_start, period_end, period_step),
    };

    fwma_batch_inner(data, &sweep, Kernel::Scalar, 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 fwma_batch_metadata_js(
    period_start: usize,
    period_end: usize,
    period_step: usize,
) -> Vec<f64> {
    fn axis_usize((start, end, step): (usize, usize, usize)) -> Vec<usize> {
        if step == 0 || start == end {
            return vec![start];
        }
        let (lo, hi) = if start <= end {
            (start, end)
        } else {
            (end, start)
        };
        (lo..=hi).step_by(step).collect()
    }

    let periods = axis_usize((period_start, period_end, period_step));
    periods.into_iter().map(|p| p as f64).collect()
}

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn fwma_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 fwma_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 fwma_into(
    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 fwma_into"));
    }

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

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

        let params = FwmaParams {
            period: Some(period),
        };
        let input = FwmaInput::from_slice(data, params);

        if in_ptr == out_ptr {
            let mut temp = vec![0.0; len];
            fwma_into_slice(&mut temp, &input, Kernel::Auto)
                .map_err(|e| JsValue::from_str(&e.to_string()))?;

            let out = std::slice::from_raw_parts_mut(out_ptr, len);
            out.copy_from_slice(&temp);
        } else {
            let out = std::slice::from_raw_parts_mut(out_ptr, len);
            fwma_into_slice(out, &input, Kernel::Auto)
                .map_err(|e| JsValue::from_str(&e.to_string()))?;
        }

        Ok(())
    }
}

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

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

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

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

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

    let js_output = FwmaBatchJsOutput {
        values: output.values,
        combos: output.combos,
        rows: output.rows,
        cols: output.cols,
    };

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

#[cfg(all(target_arch = "wasm32", feature = "wasm"))]
#[wasm_bindgen]
pub fn fwma_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 fwma_batch_into"));
    }

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

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

        let combos = expand_grid(&sweep);
        let rows = combos.len();
        let cols = len;

        let total = rows
            .checked_mul(cols)
            .ok_or(JsValue::from_str("fwma_batch_into: rows*cols overflow"))?;

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

        let first = data.iter().position(|x| !x.is_nan()).unwrap_or(0);
        let warm: Vec<usize> = combos
            .iter()
            .map(|c| first + c.period.unwrap() - 1)
            .collect();

        fill_nan_prefixes_slice(rows, cols, &warm, out);

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

        Ok(rows)
    }
}