Relation between Int and SIMD[DType.si64, 1]

[ciobaca]

In my experiments, Int seems to be a standard 64-bit signed integer.

Obviously, it represents conceptually the same information as a SIMD[DType.si64, 1] (also aliased as SI64).

I would expect the two types to be interchangeable, but this is not (currently?) the case, as illustrated by the following code:

let x : Int = 9223372036854775807
let xp : SI64 = x
print(x)
print(xp)
print(xp == x) # works likely because x is being converted into an SI64 using an implicit constructor call
#print(x == xp) # fails

Is there any way to extract the information in an SI64 as an Int?

[TZDak]

Keeping in might that I have no access to test any of this yet, here's my understanding. If you were to construct an Int object from a scalar or DType.si64 and another Int object from a scalar of DType.si16, both with the exact same scalar value, such that:

x = Int(scalar<si64>(42))
y = Int(scalar<si16>(42))

(or perhaps this can be written as something like...
x = Int(DType.si64(42))
y = Int(DType.si16(42))
...but as I said, I can't test any of this yet)

if Mojo is handling Int much like Python handles int, which seems to be the goal, then x == y should be True and x is y should also be True, although again I stress that I have no way to test this yet. My reasoning is that in Python the int type is represented internally, as I understand it, using whatever type the Python interpreter determines can most efficiently contain it. Python actually caches small integer objects for efficiency and will re-use them if the same int value is needed. Since Mojo's Int class seems to be meant as a sort of generic integer, I figure they're probably aiming for something similar. Hopefully, somebody who knows for sure will respond to you soon, but since that hasn't happened yet I thought I would give it my best try.

DAK

Edit --- With regard to your test, I think the assignment "let xp : SI64 = x" in Mojo would function as a case conversion. Perhaps try printing the type of each variable. I also wonder if repr might show something that might not be seen otherwise.