[llvm] [AMDGPU] Fix excessive stack usage in SIInsertWaitcnts::run (PR #134835)

[David Stuttard via llvm-commits]

================
@@ -2623,12 +2623,17 @@ bool SIInsertWaitcnts::run(MachineFunction &MF) {
         else
           *Brackets = *BI.Incoming;
       } else {
-        if (!Brackets)
+        if (!Brackets) {
           Brackets = std::make_unique<WaitcntBrackets>(
               ST, MaxCounter, Limits, WaitEventMaskForInst, SmemAccessCounter);
-        else
-          *Brackets = WaitcntBrackets(ST, MaxCounter, Limits,
-                                      WaitEventMaskForInst, SmemAccessCounter);
+        } else {
+          // Reinitialize in-place. N.B. do not do this by assigning from a
+          // temporary because the WaitcntBrackets class is large and it could
+          // cause this function to use an unreasonable amount of stack space.
+          Brackets->~WaitcntBrackets();
+          new (Brackets.get()) WaitcntBrackets(
----------------
dstutt wrote:

Is this placement-new syntax?

https://github.com/llvm/llvm-project/pull/134835