Struct PrefetchConfig 

pub struct PrefetchConfig {
    pub lookahead_blocks: usize,
    pub strategy: PrefetchStrategy,
}

Configuration for prefetch behavior in kernel loops.

Fields

lookahead_blocks: usize

How many blocks ahead to prefetch in the inner loop. Higher values hide more latency but consume more cache.

strategy: PrefetchStrategy

Which prefetch strategy to use.

Implementations

impl PrefetchConfig

pub fn for_gemv() -> Self

Create a config optimized for GEMV (single vector, temporal reuse of weights).

pub fn for_gemm(batch_size: usize) -> Self

Create a config optimized for GEMM (batch, streaming weights).

In GEMM, each weight block is reused across the M dimension, so temporal locality is still useful. For very large M, however, the first-touch of weight blocks benefits from non-temporal prefetch to avoid polluting L1 with data that won’t be reused for many iterations.

pub fn none() -> Self

No prefetching (baseline for benchmarking).

Trait Implementations

impl Clone for PrefetchConfig

fn clone(&self) -> PrefetchConfig

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for PrefetchConfig

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for PrefetchConfig

fn default() -> Self

Returns the “default value” for a type.