Vector256.Create messing with upper bits

[ghost]

Not seeing why this happens in C# in both .Net 5/6, but not .Net Core 3.1:

    var v = Vector256.Create(1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f);      //<1, 2, 3, 4, 5, 6, 7, 8>
    v = Vector256.Create(1.0f);                                                    //<1, 1, 1, 1, 0, 0, 0, 7.17E-43>
    v = Vector256.Create(-1.0f, -2.0f, -3.0f, -4.0f, -5.0f, -6.0f, -7.0f, -8.0f);  //<-1, -2, -3, -4, -5, -6, -7, -8>
    v = Vector256.Create(1.0f);                                                    //<1, 1, 1, 1, 1, 1, 1, 1>

The upper bits of line 2 should have matched line 4. However, moving the code around makes it occur with line 1.

[SingleAccretion]

What versions of .NET 5/6 does this reproduce on? Could you try the latest main from dotnet/installer and/or the latest .NET 5 version (5.0.7)? A bug similar to this was fixed at some point.

How does the exact reproduction look like (are you inspecting values under the debugger?)?

For reference, the following:

var v = Vector256.Create(1.0f, 2.0f, 3.0f, 4.0f, 5.0f, 6.0f, 7.0f, 8.0f);    
Console.WriteLine(v);
v = Vector256.Create(1.0f);                                                  
Console.WriteLine(v);
v = Vector256.Create(-1.0f, -2.0f, -3.0f, -4.0f, -5.0f, -6.0f, -7.0f, -8.0f);
Console.WriteLine(v);
v = Vector256.Create(1.0f);                                                  
Console.WriteLine(v);

Prints the expected values for me on 44f050a.

[SingleAccretion]

So I take it this reproduces on the latest main, but only under the debugger? Interesting.

[tannergooding]

@davmason, @tommcdon.

[davmason]

@hoyosjs have you seen this?

[hoyosjs]

No, this would be new. And the debugger has issues handling Vector already... Creating an issue... Cc:@sdmaclea as this looks like instruction decoding, which we covered when he generated look up tables.

[hoyosjs]

Opened #54025