[llvm] [AMDGPU][SIInsertWaitcnts] Do not add s_waitcnt when the counters are known to be 0 already (PR #72830)

[Jay Foad via llvm-commits]

================
@@ -886,34 +900,39 @@ bool SIInsertWaitcnts::applyPreexistingWaitcnt(
     if (II.isMetaInstruction())
       continue;
 
-    if (II.getOpcode() == AMDGPU::S_WAITCNT) {
-      // Conservatively update required wait if this waitcnt was added in an
-      // earlier pass. In this case it will not exist in the tracked waitcnt
-      // set.
-      if (!TrackedWaitcntSet.count(&II)) {
-        unsigned IEnc = II.getOperand(0).getImm();
-        AMDGPU::Waitcnt OldWait = AMDGPU::decodeWaitcnt(IV, IEnc);
-        Wait = Wait.combined(OldWait);
-      }
+    unsigned Opcode = II.getOpcode();
+    bool CanFullyDiscardWaitcntSequence = SIInstrInfo::isSoftWaitcnt(Opcode);
+
+    if (SIInstrInfo::isWaitcnt(Opcode)) {
+      // Update required wait count. If this is a soft waitcnt (= it was added
+      // by an earlier pass), it may be entirely removed.
+      unsigned IEnc = II.getOperand(0).getImm();
+      AMDGPU::Waitcnt OldWait = AMDGPU::decodeWaitcnt(IV, IEnc);
+      if (CanFullyDiscardWaitcntSequence)
+        ScoreBrackets.simplifyWaitcnt(OldWait);
+      Wait = Wait.combined(OldWait);
 
       // Merge consecutive waitcnt of the same type by erasing multiples.
-      if (!WaitcntInstr) {
+      if (!WaitcntInstr &&
+          (Wait.hasWaitExceptVsCnt() || !CanFullyDiscardWaitcntSequence)) {
         WaitcntInstr = ⅈ
       } else {
         II.eraseFromParent();
         Modified = true;
       }
----------------
jayfoad wrote:

I find this easier to understand as it has fewer negatives:
```suggestion
      if (WaitcntInstr ||
          (!Wait.hasWaitExceptVsCnt() && CanFullyDiscardWaitcntSequence)) {
        II.eraseFromParent();
        Modified = true;
      } else {
        WaitcntInstr = ⅈ
      }
```

https://github.com/llvm/llvm-project/pull/72830