[llvm-branch-commits] [llvm] AMDGPU/GlobalISel: RegBankLegalize (PR #112864)

[Petar Avramovic via llvm-branch-commits]

================
@@ -69,6 +81,241 @@ FunctionPass *llvm::createAMDGPURBLegalizePass() {
 
 using namespace AMDGPU;
 
+const RegBankLegalizeRules &getRules(const GCNSubtarget &ST,
+                                     MachineRegisterInfo &MRI) {
+  static std::mutex GlobalMutex;
+  static SmallDenseMap<unsigned, std::unique_ptr<RegBankLegalizeRules>>
+      CacheForRuleSet;
+  std::lock_guard<std::mutex> Lock(GlobalMutex);
----------------
petar-avramovic wrote:

Never changes use lock_guard+mutex, why switch to call_once?

https://github.com/llvm/llvm-project/pull/112864