Openssl fixes

[yuri91]

No description provided.

[Copilot]

Pull request overview

This PR addresses build/runtime issues (notably seen with OpenSSL) in the Cheerp-modified LLVM tree by tightening SROA legality checks, relaxing an overly strict debug assertion in registerization, and teaching identical code folding how to compare fence instructions.

Changes:

• Prevent SROA integer-widening from proceeding on non-byte-addressable targets when it would require creating illegal integer widths (e.g. i24) for mem-intrinsic slices.
• Stop VertexColorer::iterativeDeepening from asserting when called with an exhausted IterationsCounter.
• Add Instruction::Fence handling to IdenticalCodeFolding::equivalentInstruction.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 2 comments.

File	Description
llvm/lib/Transforms/Scalar/SROA.cpp	Adds a legality gate to avoid creating illegal integer slice widths in NBA (non-byte-addressable) mode.
llvm/lib/CheerpUtils/Registerize.cpp	Disables a debug assertion that can fire when the iteration budget is zero.
llvm/lib/CheerpUtils/IdenticalCodeFolding.cpp	Adds equivalence logic for fence instructions (ordering + sync scope).

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Leaving the old assertion commented out makes the intent unclear and leaves dead code behind. Prefer either removing it entirely or replacing it with an explicit guard/early return when counter.remaining() == 0 (and, if useful, a short comment explaining why 0 iterations is a valid input).

Suggested change

//assert(counter.remaining() > 0);

[Copilot]

Consider adding a regression test for the new non-byte-addressable DataLayout behavior here (e.g., a module with datalayout including 'b' and a 3-byte memintrinsic slice) to ensure SROA no longer tries integer widening that would require creating illegal integer types like i24.