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

[Austin Kerbow via llvm-branch-commits]

================
@@ -751,6 +767,9 @@ class WaitcntBrackets {
   struct VMEMInfo {
     // Scores for all instruction counters. Zero-initialized.
     CounterValueArray Scores{};
+    // Later ordered XDL writes that must remain outstanding while this
+    // VA_VDST write is pending.
+    unsigned NumOrderedXDLsAfterWrite = 0;
----------------
kerbowa wrote:

I'd rather keep it next to the scores unless you feel strongly about it. A separate map would need its own lookup on every write plus merge handling.

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