Function fused_k_norm_rope_cache_write_bf16_mrope 

pub fn fused_k_norm_rope_cache_write_bf16_mrope(
    gpu: &dyn GpuBackend,
    kernel: KernelHandle,
    k_in: DevicePtr,
    k_norm_weight: DevicePtr,
    pos_t: DevicePtr,
    pos_h: DevicePtr,
    pos_w: DevicePtr,
    k_cache: DevicePtr,
    slot_mapping: DevicePtr,
    num_tokens: u32,
    num_kv_heads: u32,
    head_dim: u32,
    rotary_dim: u32,
    block_size: u32,
    rms_eps: f32,
    theta: f32,
    stream: u64,
) -> Result<()>

MRoPE-interleaved variant — selects abs position from pos_t/pos_h/pos_w based on pair_idx % 3. For text-only inputs (pos_h == pos_w == pos_t) the result is bit-identical to the scalar-position variant.