pub enum SplitkPolicy {
Legacy,
Auto,
Pinned(u32),
}Expand description
How a target picks its paged-decode split count.
Variants§
Legacy
The pre-#928 rule: sm_count / (num_q_heads * split_ref_seqs), or 1
when that product already covers the device. Reads the runtime
reference batch, which is why it is not the default anywhere new.
Auto
Fill SPLITK_TARGET_WAVES waves at the SINGLE-STREAM shape:
clamp(ceil(WAVES * sm_count / num_q_heads), 1, MAX_DECODE_SPLITS).
Pinned(u32)
An operator-pinned count, clamped to 1..=MAX_DECODE_SPLITS. 0 and
off arrive here as Pinned(1) — one split IS no split-K.
Implementations§
Trait Implementations§
Source§impl Clone for SplitkPolicy
impl Clone for SplitkPolicy
Source§fn clone(&self) -> SplitkPolicy
fn clone(&self) -> SplitkPolicy
Returns a duplicate of the value. Read more
1.0.0 · Source§fn clone_from(&mut self, source: &Self)
fn clone_from(&mut self, source: &Self)
Performs copy-assignment from
source. Read moreSource§impl Debug for SplitkPolicy
impl Debug for SplitkPolicy
Source§impl PartialEq for SplitkPolicy
impl PartialEq for SplitkPolicy
impl Copy for SplitkPolicy
impl Eq for SplitkPolicy
impl StructuralPartialEq for SplitkPolicy
Auto Trait Implementations§
impl Freeze for SplitkPolicy
impl RefUnwindSafe for SplitkPolicy
impl Send for SplitkPolicy
impl Sync for SplitkPolicy
impl Unpin for SplitkPolicy
impl UnwindSafe for SplitkPolicy
Blanket Implementations§
Source§impl<T> BorrowMut<T> for Twhere
T: ?Sized,
impl<T> BorrowMut<T> for Twhere
T: ?Sized,
Source§fn borrow_mut(&mut self) -> &mut T
fn borrow_mut(&mut self) -> &mut T
Mutably borrows from an owned value. Read more
Source§impl<T> CloneToUninit for Twhere
T: Clone,
impl<T> CloneToUninit for Twhere
T: Clone,
§impl<Q, K> Equivalent<K> for Q
impl<Q, K> Equivalent<K> for Q
§fn equivalent(&self, key: &K) -> bool
fn equivalent(&self, key: &K) -> bool
Compare self to
key and return true if they are equal.