[clang] [llvm] [AMDGPU] Introduce asyncmark/wait intrinsics (PR #180467)

[Sameer Sahasrabuddhe via cfe-commits]

================
@@ -1220,20 +1252,43 @@ void WaitcntBrackets::updateByEvent(WaitEventType E, MachineInstr &Inst) {
         setVMemScore(LDSDMA_BEGIN + Slot, T, CurrScore);
     }
 
+    // FIXME: Not supported on GFX12 yet. Newer async operations use other
+    // counters too, so will need a map from instruction or event types to
+    // counter types.
+    if (Context->isAsyncLdsDmaWrite(Inst) && T == LOAD_CNT) {
+      assert(!SIInstrInfo::usesASYNC_CNT(Inst));
----------------
ssahasra wrote:

Done in #181095

https://github.com/llvm/llvm-project/pull/180467