pub fn grouped_gemm_mla(
gpu: &dyn GpuBackend,
kernel: KernelHandle,
a: DevicePtr,
b: DevicePtr,
c: DevicePtr,
m: u32,
g: u32,
k_g: u32,
n_g: u32,
a_stride: u32,
c_stride: u32,
stream: u64,
) -> Result<()>Expand description
2026-09-25: g independent GEMMs C_g[m, n_g] = A_g[m, k_g] @ B_g[n_g, k_g]^T
in one launch. A_g sits at column g * k_g of A, B_g at row g * n_g of
B, and C_g at column g * n_g of C; one block per (row, group) computes
4 outputs.