Vectorized implementation of Log/Pow/Cbrt/RootN is broken for some inputs

[stephentoub]

(Reported by @michaelgsharp)

Repro:

using System.Numerics.Tensors;

float[] dest = new float[4];

// Non-vectorized
float[] values = [-1, -2, -3];
TensorPrimitives.Pow(values, 3f, dest);
Console.WriteLine($"{dest[0]} {dest[1]} {dest[2]}");

// Vectorized
values = [-1, -2, -3, -4];
TensorPrimitives.Pow(values, 3f, dest);
Console.WriteLine($"{dest[0]} {dest[1]} {dest[2]} {dest[3]}");

prints

-1 -8 -27
NaN NaN NaN NaN

[dotnet-policy-service]

Tagging subscribers to this area: @dotnet/area-system-numerics-tensors
See info in area-owners.md if you want to be subscribed.

[stephentoub]

This extends beyond Pow. We have bugs here for Cbrt, Log, Pow, and RootN.

[stephentoub]

Closing as this was disabled in the implementation.