Struct BatchedGemmArgs 

pub struct BatchedGemmArgs<'a, T>
where
    T: Element,
{
    pub a: MatrixRef<'a, T>,
    pub stride_a: i64,
    pub b: MatrixRef<'a, T>,
    pub stride_b: i64,
    pub c: Option<MatrixRef<'a, T>>,
    pub stride_c: i64,
    pub d: MatrixMut<'a, T>,
    pub stride_d: i64,
    pub alpha: <T as Element>::Scalar,
    pub beta: <T as Element>::Scalar,
}

Per-launch arguments for a BatchedGemmPlan::run call.

stride_* fields are in elements, not bytes — matching CUTLASS’s GemmBatched API. Pass 0 for stride if the same matrix should be reused across all batches (broadcast).

Fields

a: MatrixRef<'a, T>

Left input — base pointer for batch 0.

stride_a: i64

Element offset between consecutive A batches.

b: MatrixRef<'a, T>

Right input — base pointer for batch 0.

stride_b: i64

Element offset between consecutive B batches.

c: Option<MatrixRef<'a, T>>

Optional accumulation source.

stride_c: i64

Element offset between consecutive C batches. Ignored when c is None.

d: MatrixMut<'a, T>

Output — base pointer for batch 0.

stride_d: i64

Element offset between consecutive D batches.

alpha: <T as Element>::Scalar

α multiplier (shared across batches). Scalar type matches T::Scalar — f32 for f16/bf16/f32/F32Strict, f64 for f64.

beta: <T as Element>::Scalar

β multiplier (shared across batches). Forced to 0 internally when c is None.

Trait Implementations

impl<'a, T> Debug for BatchedGemmArgs<'a, T>

where T: Debug + Element, <T as Element>::Scalar: Debug,

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

Formats the value using the given formatter.