Function moe_w4a16_grouped_gemm_ptrtable_m256 

pub fn moe_w4a16_grouped_gemm_ptrtable_m256(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    a: DevicePtr,
    b_packed_ptrs: DevicePtr,
    b_scale_ptrs: DevicePtr,
    scale2_vals: DevicePtr,
    c: DevicePtr,
    expert_offsets: DevicePtr,
    sorted_token_ids: DevicePtr,
    num_experts: u32,
    n_out: u32,
    k: u32,
    max_m_tiles: u32,
    stream: u64,
) -> Result<()>

moe_w4a16_grouped_gemm_ptrtable with M_TILE=256 (512-thread block, 16 warps). Caller must pass max_m_tiles computed against 256, not 64 — see the div_ceil(4) at the call site, mirroring the m128 variant’s div_ceil(2).

DEFAULT-OFF, measured non-win — ~20% slower per call than the base kernel end-to-end (31.30 vs 26.17 ms avg under nsys). See launch_grouped_gemm.