[llvm] [NFC][AMDGPU] Compute always reserved registers once (PR #132006)

[Jay Foad via llvm-commits]

================
@@ -629,12 +629,16 @@ SIRegisterInfo::getMaxNumVectorRegs(const MachineFunction &MF) const {
   return std::pair(MaxNumVGPRs, MaxNumAGPRs);
 }
 
-BitVector SIRegisterInfo::getReservedRegs(const MachineFunction &MF) const {
-  BitVector Reserved(getNumRegs());
+BitVector SIRegisterInfo::getAlwaysReservedRegs() const {
+  static BitVector AlwaysReservedRegs;
+  // Already been calculated, so do not compute again.
+  if (AlwaysReservedRegs.size() == getNumRegs())
----------------
jayfoad wrote:

Doesn't this need to be done in a thread-safe way??

https://github.com/llvm/llvm-project/pull/132006