Mark MemoryMarshal.Cast methods for aggressive inlining

[AndyAyersMS]

Inlining doesn't streamline the cast logic any, but it facilitates
caller struct promotion which can substantially boost perf.

See discussion in dotnet/corefx#27485.

[AndyAyersMS]

@ianhays PTAL

[mikedn]

Hmm, I really should improve that cast code to avoid bloating the caller...

[ahsonkhan]

but it facilitates
caller struct promotion which can substantially boost perf

What other Span APIs would this be applicable to?

[AndyAyersMS]

Perf impact measured locally...

[AndyAyersMS]

If you have a Span local or parameter and inline all methods that it is passed to, then typically that local or parameter can be promoted. If there's just one method that doesn't get inlined it will block promotion.

So in the case of Marvin.ComputeHash the data parameter can only be promoted if all the calls to Cast are inlined, even if those happen in rarely taken paths.

[ahsonkhan]

If you have a Span local or parameter and inline all methods that it is passed to, then typically that local or parameter can be promoted. If there's just one method that doesn't get inlined it will block promotion.

Should we mark GetReference as AggressiveInlining?
https://github.com/dotnet/corefx/blob/master/src/Common/src/CoreLib/System/Runtime/InteropServices/MemoryMarshal.Fast.cs#L34
https://github.com/dotnet/corefx/blob/master/src/System.Memory/src/System/Runtime/InteropServices/MemoryMarshal.Portable.cs#L31

For cases like these:
https://github.com/dotnet/corefx/blob/40cbd5636851093309b841db4d53af369b7ef1c6/src/System.IO.FileSystem/src/System/IO/CharSpanExtensions.Unix.cs#L20
https://github.com/dotnet/corefx/blob/40cbd5636851093309b841db4d53af369b7ef1c6/src/Common/src/Interop/Unix/System.Security.Cryptography.Native/Interop.EcDsa.cs#L20

[jkotas]

GetReference is small enough to be naturally inlinable. You do not have to force it.