Skip to main content

Module gemm

Module gemm 

Source
Expand description

Typed GemmRequest<T> + GemmDispatch impls.

cudarc 0.19 exposes cudarc::cublas::Gemm<T> for f32, f64, and (under feature f16) half::f16 and half::bf16. The dispatcher re-uses that safe trait so we don’t have to touch cublasGemmEx directly for the common dtypes — fp8 is the future follow-up that lights up crate::sys::cublas::gemm_ex once the cublas-fp8 feature is wired (see super::scaling).

Structs§

GemmRequest
Typed cuBLAS gemm request: C = α·op(A)·op(B) + β·C.