[llvm] [AMDGPU] Model ordered XDL writes in expert scheduling (PR #218326)

[via llvm-commits]

================
@@ -1344,9 +1374,9 @@ void WaitcntBrackets::purgeEmptyTrackingData() {
   SGPRs.remove_if([](const auto &P) { return P.second.empty(); });
 }
 
-void WaitcntBrackets::determineWaitForScore(AMDGPU::InstCounterType T,
-                                            unsigned ScoreToWait,
-                                            AMDGPU::Waitcnt &Wait) const {
+void WaitcntBrackets::determineWaitForScore(
+    AMDGPU::InstCounterType T, unsigned ScoreToWait, AMDGPU::Waitcnt &Wait,
+    unsigned IndependentWaitBound) const {
----------------
vporpo wrote:

This function is already too complicated for what it does. Can't you just get the waits and adjust it externally at the call site?

https://github.com/llvm/llvm-project/pull/218326