[llvm] [X86] Prefer `lock or` over mfence (PR #106555)

[Valentin Churavy via llvm-commits]

================
@@ -754,6 +754,10 @@ def TuningUseGLMDivSqrtCosts
 def TuningBranchHint: SubtargetFeature<"branch-hint", "HasBranchHint", "true",
                                         "Target has branch hint feature">;
 
+def TuningAvoidMFENCE
----------------
vchuravy wrote:

I am sympathetic to that, but also following GCCs lead here.

https://github.com/llvm/llvm-project/pull/106555