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

[Simon Pilgrim via llvm-commits]

================
@@ -882,7 +886,8 @@ def ProcessorFeatures {
   list<SubtargetFeature> NHMTuning = [TuningMacroFusion,
                                       TuningSlowDivide64,
                                       TuningInsertVZEROUPPER,
-                                      TuningNoDomainDelayMov];
+                                      TuningNoDomainDelayMov,
+                                      TuningAvoidMFENCE];
----------------
RKSimon wrote:

If we do that we might as well support it on any 64-bit capable CPU

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