expensive readonly HashSet<T> construction

[kasperk81]

https://godbolt.org/z/eE5svxTMo

class A benefits from JIT optimizations: the array setup is simplified to a direct allocation and pointer assignments, making it efficient with minimal overhead.

In class B, using HashSet<string> is more costly:

• The JIT initializes hashing and comparison structures and performs helper calls (e.g., AddIfNotPresent).
• Each call involves hash calculations and duplicate checks, which the JIT cannot eliminate, leading to higher overhead.

@EgorBo Could the JIT optimize HashSet initialization for known unique values, reducing redundant checks?

[huoyaoyuan]

class A benefits from JIT optimizations
Could the JIT optimize HashSet initialization for known unique values, reducing redundant checks?

No, because it's not done by the JIT and out of scope.

The work for array is done by C# compiler: https://sharplab.io/#v2:D4AQDABCCMAsDcBYAUCkAmCBBFBvFEhUAzFNGANoC6EAshALwB8BRFrRRARAIYBGAYy4AaDpwhcAHgE8AXiLGcu0dMS5iqSZAF8UaTACE8YkKQASPAM4ALAMoBTAC4AeGGCZ1GLZJ3Y/xhLyCCv4BUnIhAYEqahpaushAA==
JIT is just compiling the IL as-is.

There is optimal building support for FrozenSet<T> (#102223), but not HashSet<T>.