Incorrect @code_typed for a simple program about julia

[vtjnash]

%51 equals to %47

%51 is never equal to %47. Phi nodes apply simultaneously, not sequentially

[thautwarm]

P.S: if we prevent view from being inlined, the result gets correct again.

# replace `view` with `view2`
@noinline view2(args...) = view(args...)

[vtjnash]

it shows %47 is the local variable n

That is not quite how values work. The variable is gone, replaced by zero or more % values that represent the value it had at various points in the code. In particular, it looks like %51 is a time-shifted copy of %47 (e.g. it always has the value from the previous iteration). Can you say a bit more about what result you were expecting? I get this, which looks correct at a glance (that is, I see @interpret simu!(U, u) at least populate the same values into U):

julia> U = reshape(1:3*3*3, (3,3,3)) |> collect;

julia> u = reshape(10 .+ (1:3*3), (3,3)) |> collect;

julia> simu!(U, u)
1
2

julia> U
3×3×3 Array{Int64, 3}:
[:, :, 1] =
 1   4  7
 2  45  8
 3  47  9

[:, :, 2] =
 10  13  16
 11  49  17
 12  92  18

[:, :, 3] =
 19  22  25
 20  23  26
 21  24  27

[thautwarm]

That is not quite how values work. The variable is gone, replaced by zero or more % values that represent the value it had at various points in the code. In particular, it looks like %51 is a time-shifted copy of %47

Yes, my point is that %51 equals to %47 which actually stores an integer 2 when n = 2.

Can you say a bit more about what result you were expecting?
I expect the first arrayref is not using %51 (%47, or n) as the its 5-th argument but a value extracted from some structure.

Given size(U) = (3, 3, 3), when n=2, the 5-th argument of arrayref should be 1 instead of 2 at runtime.

I see @InterPret simu!(U, u) at least populate the same values

(EIDT) I don't think @interpret is related because the interpreter uses the untyped form.

Julia does run the MVE code correctly producing correct results, but I don't think the result of @code_typed is correct.

In my derivation, if U = ones(3, 3, 3) is given, and using the following inputs, the @code_typed output is different from the real execution.

U = ones(3, 3, 3)
u = ones(3, 3)
simu!(U, u)
display(U[:, :, 2])

# julia/correct result
#=
[:, :, 2] =
 1.0  1.0  1.0
 1.0  3.0  1.0
 1.0  4.0  1.0
=#
# according to code_typed
#=
[:, :, 2] =
 1.0  1.0  1.0
 1.0  2.0  1.0
 1.0  3.0  1.0
=#

[vtjnash]

If Julia runs the code correctly, then it seems more likely you are interpreting it wrong mentally than that the code is wrong but still gives the right answer.

[thautwarm]

Thanks a lot for your information, I can confirm this is not a bug.
Very sorry for taking your time with this issue!