Why should width be explicitly mentioned in SIMD

[alfie-nsugh]

In this code

from tensor import Tensor, TensorSpec, TensorShape
def main():
    alias type = DType.int64
    var spec = TensorSpec(type,5)
    var s = Tensor[type](spec)
    var t = Tensor[type](spec)
    var u = Tensor[type](spec)
    var slist = List(1,2,3,4,5)
    var tlist = List(2,3,5,7,11)
    for i in range(5):
        s[i] = slist[i]
        t[i] = tlist[i]
    u.store[5](0,s.load[width=5](0))
    print(u)

I see that the result is: Tensor([[1, 2, 3, 4, 5]], dtype=int64, shape=5) which is what I want.

But I get for

def main():
    alias type = DType.int64
    var spec = TensorSpec(type,5)
    var s = Tensor[type](spec)
    var t = Tensor[type](spec)
    var u = Tensor[type](spec)
    var slist = List(1,2,3,4,5)
    var tlist = List(2,3,5,7,11)
    for i in range(5):
        s[i] = slist[i]
        t[i] = tlist[i]
    u.store[5](0,s.load[5](0))
    print(u)

This result
Tensor([[1, 0, 0, 0, 0]], dtype=int64, shape=5)

It took quite a while to figure that out. Is this a bug or does it make sense that we need to be this explicit with the width?

[rarebreed]

I would call that a bug.

Mojo appears to be calling the overloaded load method of load[len: Int, width: Int](self: Self, indices: StaticIntTuple[len]) -> SIMD[dtype, width] which it should not.

According to the manual's rules:

1. Candidates with the minimal number of implicit conversions (in both arguments and parameters).
2. Candidates without variadic arguments.
3. Candidates without variadic parameters.
4. Candidates with the shortest parameter signature.
5. Non-@staticmethod candidates (over @staticmethod ones, if available).

The load method of load[width: Int](self: Self, index: Int) -> SIMD[dtype, width] should have been called because rule 1 should have been applied.

[rarebreed]

And actually, it looks like the store method is also calling the wrong overloaded method too.