Function moe_unpermute_blend 

pub fn moe_unpermute_blend(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    expert_output: DevicePtr,
    output: DevicePtr,
    token_to_perm: DevicePtr,
    topk_weights: DevicePtr,
    shared_out: DevicePtr,
    normed: DevicePtr,
    gate_weight: DevicePtr,
    hidden_size: u32,
    num_tokens: u32,
    topk: u32,
    stream: u64,
) -> Result<()>

2026-09-28: moe_unpermute_reduce_indexed then moe_batched_blend in one launch (kernel moe_unpermute_blend, kernels/gb10/common/moe_unpermute_blend.cu), with the same output bits; one 256-thread block per token. Needs hidden_size % 8 == 0.