Struct RaspBatch 

pub struct RaspBatch<T> { /* private fields */ }

Structure-of-arrays storage for bounds-checked pointers.

T is a phantom type; each entry refers to an externally-owned [T] region. The caller is responsible for keeping the regions alive for the lifetime of the batch. Use push_from_slice with the returned slice as a borrow anchor.

Implementations

impl<T> RaspBatch<T>

pub fn new() -> Self

pub fn with_capacity(n: usize) -> Self

pub fn len(&self) -> usize

pub fn is_empty(&self) -> bool

pub fn capacity(&self) -> usize

pub fn push_from_slice<'a>( &mut self, slice: &'a [T], perms: u32, ) -> Result<(RaspBatchIndex<T>, &'a [T]), RaspError>

Push a new pointer from a borrowed slice. The returned slice is the lifetime anchor; the batch’s pointer is valid only as long as the anchor is held.

pub fn push_raw( &mut self, ptr: u64, base: u64, length: u32, perms: u32, ) -> Result<RaspBatchIndex<T>, RaspError>

Push from raw integer fields (caller manages target lifetime).

pub fn check_read_scalar(&self, idx: RaspBatchIndex<T>) -> Result<(), RaspError>

Per-element scalar check. Used by per-index call sites and as the correctness oracle for the SIMD batch path.

pub fn raw_ptr(&self, idx: RaspBatchIndex<T>) -> Option<*const T>

Raw pointer at the given index. Returns None for an out-of-range index. The pointer is NOT validated; pair with check_read_scalar (or use read_at which combines both).

pub unsafe fn read_at(&self, idx: RaspBatchIndex<T>) -> Result<T, RaspError>

where T: Copy,

Bounds + permission check followed by a dereference. The caller’s responsibility for the original target lifetime still applies (see push_from_slice’s borrow anchor).

Safety

The target of the pointer at idx must still be valid (alive, properly aligned for T, and accessible via the permissions encoded). The check enforces the permissions recorded at push time; it does NOT prove the target’s allocation has not been freed.

pub fn check_read_all_scalar(&self) -> Vec<Result<(), RaspError>>

Batch scalar check across the whole array. Reference path.

pub fn count_valid_scalar(&self) -> u32

Count of valid entries (Ok results) via scalar path. Avoids the Vec<Result> allocation when only the count is needed.

pub unsafe fn count_valid_avx2(&self) -> u32

AVX2-accelerated count of valid entries. Processes 4 elements per loop iteration using contiguous SIMD loads from each parallel Vec.

Safety

Caller must guarantee AVX2 is supported by the runtime CPU.

pub unsafe fn count_valid_avx512(&self) -> u32

AVX-512F count of valid entries. Processes 8 elements per iteration: ZMM (8x u64) loads for ptrs/bases, YMM (8x u32) loads for lengths/perms, mask-producing _mm512_cmpgt_epi64_mask for both bounds checks. Doubles the per-iteration throughput of the AVX2 path.

Safety

Caller must guarantee AVX-512F is supported by the runtime CPU.

pub fn count_valid(&self) -> u32

Cross-platform safe count of valid entries. Dispatches to count_valid_avx512 when AVX-512F is present, then count_valid_avx2, then count_valid_scalar.

pub unsafe fn check_read_all_avx2(&self) -> Vec<Result<(), RaspError>>

AVX2-accelerated full validation that writes per-index results. Slower than count_valid_avx2 because it materialises a Vec<Result> instead of just counting; use this when the caller needs per-index error attribution.

Safety

Caller must guarantee AVX2 is supported by the runtime CPU.

pub unsafe fn check_read_all_avx512(&self) -> Vec<Result<(), RaspError>>

AVX-512F full validation that writes per-index results. Processes 8 elements per iteration via ZMM loads and mask-producing _mm512_cmpgt_epi64_mask. Bit-exact equivalent of check_read_all_avx2 and check_read_all_scalar.

Safety

Caller must guarantee AVX-512F is supported by the runtime CPU.

pub fn check_read_all(&self) -> Vec<Result<(), RaspError>>

Cross-platform safe full validation.

Trait Implementations

impl<T> Default for RaspBatch<T>

fn default() -> Self

Returns the “default value” for a type.

impl<T: Send> Send for RaspBatch<T>

impl<T: Sync> Sync for RaspBatch<T>