spiral-rs-fork/spiral-rs/src/aligned_memory.rs

use std::{alloc::{alloc_zeroed, dealloc, Layout}, slice::{from_raw_parts, from_raw_parts_mut}, ops::{Index, IndexMut}, mem::size_of};


const ALIGN_SIMD: usize = 64; // enough to support AVX-512
pub type AlignedMemory64 = AlignedMemory<ALIGN_SIMD>;

pub struct AlignedMemory<const ALIGN: usize> {
    p: *mut u64,
    sz_u64: usize,
    layout: Layout
}

impl<const ALIGN: usize> AlignedMemory<{ALIGN}> {
    pub fn new(sz_u64: usize) -> Self {
        let sz_bytes = sz_u64 * size_of::<u64>();
        let layout = Layout::from_size_align(sz_bytes, ALIGN).unwrap();

        let ptr;
        unsafe {
            ptr = alloc_zeroed(layout);
        }

        Self {
            p: ptr as *mut u64,
            sz_u64,
            layout
        }
    }

    pub fn as_slice(&self) -> &[u64] {
        unsafe {
            from_raw_parts(self.p, self.sz_u64)
        }
    }

    pub fn as_mut_slice(&mut self) -> &mut [u64] {
        unsafe {
            from_raw_parts_mut(self.p, self.sz_u64)
        }
    }

    pub fn len(&self) -> usize {
        self.sz_u64
    }
}

impl<const ALIGN: usize> Drop for AlignedMemory<{ALIGN}> {
    fn drop(&mut self) {
        unsafe {
            dealloc(self.p as *mut u8, self.layout);
        }
    }
}

impl<const ALIGN: usize> Index<usize> for AlignedMemory<{ALIGN}> {
    type Output = u64;

    fn index(&self, index: usize) -> &Self::Output {
        &self.as_slice()[index]
    }
}

impl<const ALIGN: usize> IndexMut<usize> for AlignedMemory<{ALIGN}> {
    fn index_mut(&mut self, index: usize) -> &mut Self::Output {
        &mut self.as_mut_slice()[index]
    }
}

impl<const ALIGN: usize> Clone for AlignedMemory<{ALIGN}> {
    fn clone(&self) -> Self {
        let mut out = Self::new(self.sz_u64);
        out.as_mut_slice().copy_from_slice(self.as_slice());
        out
    }
}