launch_k3_moe_e8m0_ptrtable

Function launch_k3_moe_e8m0_ptrtable 

Source
pub fn launch_k3_moe_e8m0_ptrtable(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    a: DevicePtr,
    packed: &[QuantizedWeight],
    c: DevicePtr,
    expert_offsets: DevicePtr,
    sorted_token_ids: DevicePtr,
    num_experts: u32,
    n_out: u32,
    k: u32,
    stream: u64,
) -> Result<()>
Expand description

One grouped E8M0 GEMM: C[M, N] = A[M, K] @ B_packed[N, K/2] via ptrtable.

Each expert in packed owns one output row (expert_offsets 0..=num_experts).