[llvm] [AMDGPU][VOPD] Limit VOPDPairing from reaching over load dependencies (PR #201930)

[Joe Nash via llvm-commits]

================
@@ -309,6 +339,38 @@ struct VOPDPairingMutation : ScheduleDAGMutation {
         if (!hasLessThanNumFused(*JSUI, 2) ||
             !shouldScheduleAdjacent(TII, ST, IMI, *JMI))
           continue;
+
+        bool LoadsMayOverlap = !ILoadSuccs.empty() && [&] {
+          BitVector JVisited(
----------------
Sisyph wrote:

Use one scratch bitvector reassigned on each iteration of J to avoid reallocation

https://github.com/llvm/llvm-project/pull/201930