Mojo function

grouped_matmul_sm90

grouped_matmul_sm90[c_type: DType, c_shape: DimList, a_type: DType, a_shape: DimList, b_type: DType, b_shape: DimList, //, *, transpose_b: Bool = True, wgmma_shape: IndexList[3] = Index(64, 256, 16), config: MatmulConfig[a_type, b_type, c_type, transpose_b, wgmma_shape] = default_config_sm90[::DType,::DType,::DType,::Bool,::IndexList[::Int(), elementwise_lambda_fn: OptionalReg[fn[DType, Int, Int](IndexList[2], SIMD[$0, $1]) capturing -> None] = OptionalReg[fn[DType, Int, Int](IndexList[2], SIMD[$0, $1]) capturing -> None]({:i1 0, 1})](c: NDBuffer[c_type, 2, MutableAnyOrigin, c_shape], a: NDBuffer[a_type, 2, MutableAnyOrigin, a_shape], a_offsets: NDBuffer[uint32, 1, MutableAnyOrigin], max_num_tokens_per_expert: Int, b: NDBuffer[b_type, 3, MutableAnyOrigin, b_shape], expert_ids: NDBuffer[uint32, 1, MutableAnyOrigin], num_active_experts: Int, ctx: DeviceContext)