julia> @finch_code begin
        y .= 0
        for j = parallel(_)
            for i = _
                y[j] += x[i] * A[walk(i), j]
            end
        end
    end
quote
    y_lvl = ((ex.bodies[1]).bodies[1]).tns.bind.lvl
    y_lvl_2 = y_lvl.lvl
    y_lvl_2_locks = y_lvl_2.locks
    y_lvl_3 = y_lvl_2.lvl
    y_lvl_3_val = y_lvl_3.val
    x_lvl = (((ex.bodies[1]).bodies[2]).body.body.rhs.args[1]).tns.bind.lvl
    x_lvl_stop = x_lvl.shape
    x_lvl_2 = x_lvl.lvl
    x_lvl_2_val = x_lvl_2.val
    A_lvl = (((ex.bodies[1]).bodies[2]).body.body.rhs.args[2]).tns.bind.lvl
    A_lvl_stop = A_lvl.shape
    A_lvl_2 = A_lvl.lvl
    A_lvl_2_ptr = A_lvl_2.ptr
    A_lvl_2_idx = A_lvl_2.idx
    A_lvl_2_stop = A_lvl_2.shape
    A_lvl_3 = A_lvl_2.lvl
    A_lvl_3_val = A_lvl_3.val
    n = (Base.Threads).nthreads()
    A_lvl_2_stop == x_lvl_stop || throw(DimensionMismatch("mismatched dimension limits ($(A_lvl_2_stop) != $(x_lvl_stop))"))
    Galley.resize_if_smaller!(y_lvl_2_locks, A_lvl_stop)
    @inbounds for idx = 1:A_lvl_stop
            y_lvl_2_locks[idx] = Galley.make_lock(eltype(Vector{Base.Threads.SpinLock}))
        end
    Galley.resize_if_smaller!(y_lvl_3_val, A_lvl_stop)
    Galley.fill_range!(y_lvl_3_val, 0, 1, A_lvl_stop)
    y_lvl_3_val_2 = (Galley).transfer((Galley.CPUSharedMemory)(Galley.CPU{:id}(n)), y_lvl_3_val)
    y_lvl_2_locks_2 = (Galley).transfer((Galley.CPUSharedMemory)(Galley.CPU{:id}(n)), y_lvl_2_locks)
    A_lvl_3_val_2 = (Galley).transfer((Galley.CPUSharedMemory)(Galley.CPU{:id}(n)), A_lvl_3_val)
    A_lvl_2_ptr_2 = (Galley).transfer((Galley.CPUSharedMemory)(Galley.CPU{:id}(n)), A_lvl_2_ptr)
    A_lvl_2_idx_2 = (Galley).transfer((Galley.CPUSharedMemory)(Galley.CPU{:id}(n)), A_lvl_2_idx)
    x_lvl_2_val_2 = (Galley).transfer((Galley.CPUSharedMemory)(Galley.CPU{:id}(n)), x_lvl_2_val)
    Galley.@barrier Threads.@threads(:dynamic, for tid = 1:n
                @inbounds @fastmath(begin
                            y_lvl_3_val_3 = (Galley).transfer((Galley.CPUThread)(tid, Galley.CPU{:id}(n), (Galley.SerialTask)()), y_lvl_3_val_2)
                            y_lvl_2_locks_3 = (Galley).transfer((Galley.CPUThread)(tid, Galley.CPU{:id}(n), (Galley.SerialTask)()), y_lvl_2_locks_2)
                            A_lvl_3_val_3 = (Galley).transfer((Galley.CPUThread)(tid, Galley.CPU{:id}(n), (Galley.SerialTask)()), A_lvl_3_val_2)
                            A_lvl_2_ptr_3 = (Galley).transfer((Galley.CPUThread)(tid, Galley.CPU{:id}(n), (Galley.SerialTask)()), A_lvl_2_ptr_2)
                            A_lvl_2_idx_3 = (Galley).transfer((Galley.CPUThread)(tid, Galley.CPU{:id}(n), (Galley.SerialTask)()), A_lvl_2_idx_2)
                            x_lvl_2_val_3 = (Galley).transfer((Galley.CPUThread)(tid, Galley.CPU{:id}(n), (Galley.SerialTask)()), x_lvl_2_val_2)
                            phase_start_2 = max(1, 1 + fld(A_lvl_stop * (tid + -1), n))
                            phase_stop_2 = min(A_lvl_stop, fld(A_lvl_stop * tid, n))
                            if phase_stop_2 >= phase_start_2
                                for j_6 = phase_start_2:phase_stop_2
                                    y_lvl_q = (1 - 1) * A_lvl_stop + j_6
                                    A_lvl_q = (1 - 1) * A_lvl_stop + j_6
                                    A_lvl_2_q = A_lvl_2_ptr_3[A_lvl_q]
                                    A_lvl_2_q_stop = A_lvl_2_ptr_3[A_lvl_q + 1]
                                    if A_lvl_2_q < A_lvl_2_q_stop
                                        A_lvl_2_i1 = A_lvl_2_idx_3[A_lvl_2_q_stop - 1]
                                    else
                                        A_lvl_2_i1 = 0
                                    end
                                    phase_stop_3 = min(A_lvl_2_stop, A_lvl_2_i1)
                                    if phase_stop_3 >= 1
                                        if A_lvl_2_idx_3[A_lvl_2_q] < 1
                                            A_lvl_2_q = Galley.scansearch(A_lvl_2_idx_3, 1, A_lvl_2_q, A_lvl_2_q_stop - 1)
                                        end
                                        while true
                                            A_lvl_2_i = A_lvl_2_idx_3[A_lvl_2_q]
                                            if A_lvl_2_i < phase_stop_3
                                                A_lvl_3_val_4 = A_lvl_3_val_3[A_lvl_2_q]
                                                x_lvl_q = (1 - 1) * x_lvl_stop + A_lvl_2_i
                                                x_lvl_2_val_4 = x_lvl_2_val_3[x_lvl_q]
                                                y_lvl_2atomicArraysAcc = Galley.get_lock(Galley.CPU{:id}(n), y_lvl_2_locks_3, y_lvl_q, eltype(Vector{Base.Threads.SpinLock}))
                                                Galley.aquire_lock!(Galley.CPU{:id}(n), y_lvl_2atomicArraysAcc)
                                                y_lvl_3_val_3[y_lvl_q] = A_lvl_3_val_4 * x_lvl_2_val_4 + y_lvl_3_val_3[y_lvl_q]
                                                Galley.release_lock!(Galley.CPU{:id}(n), y_lvl_2atomicArraysAcc)
                                                A_lvl_2_q += 1
                                            else
                                                phase_stop_5 = min(phase_stop_3, A_lvl_2_i)
                                                if A_lvl_2_i == phase_stop_5
                                                    A_lvl_3_val_4 = A_lvl_3_val_3[A_lvl_2_q]
                                                    x_lvl_q = (1 - 1) * x_lvl_stop + phase_stop_5
                                                    x_lvl_2_val_5 = x_lvl_2_val_3[x_lvl_q]
                                                    y_lvl_2atomicArraysAcc_2 = Galley.get_lock(Galley.CPU{:id}(n), y_lvl_2_locks_3, y_lvl_q, eltype(Vector{Base.Threads.SpinLock}))
                                                    Galley.aquire_lock!(Galley.CPU{:id}(n), y_lvl_2atomicArraysAcc_2)
                                                    y_lvl_3_val_3[y_lvl_q] += A_lvl_3_val_4 * x_lvl_2_val_5
                                                    Galley.release_lock!(Galley.CPU{:id}(n), y_lvl_2atomicArraysAcc_2)
                                                    A_lvl_2_q += 1
                                                end
                                                break
                                            end
                                        end
                                    end
                                end
                            end
                            phase_start_6 = max(1, 1 + fld(A_lvl_stop * tid, n))
                            if A_lvl_stop >= phase_start_6
                                A_lvl_stop + 1
                            end
                        end)
            end)
    resize!(y_lvl_2_locks_2, A_lvl_stop)
    Galley.resize_if_larger!(y_lvl_3_val_2, A_lvl_stop)
    (y = Tensor((DenseLevel){Int64}((MutexLevel){Vector{Base.Threads.SpinLock}, ElementLevel{0, Int64, Int64, Vector{Int64}}}(ElementLevel{0, Int64, Int64}(y_lvl_3_val_2), y_lvl_2_locks_2), A_lvl_stop)),)
end
julia> @finch begin
        y .= 0
        for i = parallel(_)
            for j = _
                y[j] += x[i] * A[walk(i), j]
            end
        end
    end
(y = Tensor(Dense{Int64}(Mutex(Element{0, Int64, Int64}([4, 6]), [Base.Threads.SpinLock(0), Base.Threads.SpinLock(0)]), 2)),)

