Ensure that the Hardware Intrinsics feature is documented for 3.0

[tannergooding]

We should ensure that the hardware intrinsics feature, namely the types in the System.Runtime.Intrinsics namespace, is properly documented for 3.0

At minimum, this likely requires some cleanup of the existing documentation comments (which largely just reference the underlying instruction and the corresponding native intrinsic) and we may want to consider linking to the "Intel Intrinsics Guide" (similarly to what the MSVC Intrinsics documentation does).

[tannergooding]

CC. @danmosemsft, @CarolEidt, @fiigii

[fiigii]

we may want to consider linking to the "Intel Intrinsics Guide" (similarly to what the MSVC Intrinsics documentation does).

Yes, that would be great. Shall we add the links in the XML comments of API source code? If yes, do you have an existing example?

[tannergooding]

I believe that the link needs to be an aka.ms link so we can have it redirect if the URL ever changes. Otherwise, I believe we can just add it as needed.

[fiigii]

I believe that the link needs to be an aka.ms link so we can have it redirect if the URL ever changes.

I have contacted the team that is responsible for Intel Intrinsics Guide, and they said the URLs won't be changed in the future.

[ganchito55]

IMHO at least a link to https://software.intel.com/sites/landingpage/IntrinsicsGuide/ should be added at the end/beginning of the documentation. However it would be good if each instruction had each own link.

[tannergooding]

This is being worked on and I hope to have this resolved relatively soon.

[ganchito55]

I have created an example of usage with a benchmark (SIMD vs classic implementation vs LINQ implementation) that maybe could help you to improve the documentation.
https://github.com/ganchito55/simd

[tannergooding]

Duplicate of dotnet/runtime#81753

This is a legal problem. The documentation for the instructions, including the relevant psuedo-code describing how they operate, is copyrighted by the respective companies (Intel for Intel's docs, AMD for AMD's docs, etc). Due to the scale, explicit permissions are required to reproduce such documentation in our own.

There is then the additional consideration that our docs are normally creative commons, but this subset would likely not be, which adds its own complications to the process.

The current documentation does what we can and gives enough information for developers to refer to the architecture specific instruction manuals or documentation for more information on behavior.