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

[Pierre van Houtryve via llvm-commits]

================
@@ -870,6 +875,15 @@ static bool updateOperandIfDifferent(MachineInstr &MI, uint16_t OpName,
   return true;
 }
 
+bool SIInsertWaitcnts::updateWaitcntIfSoft(MachineInstr *Waitcnt) const {
+  unsigned Opcode = Waitcnt->getOpcode();
+  if (!SIInstrInfo::isSoftWaitcnt(Opcode))
+    return false;
+
+  Waitcnt->setDesc(TII->get(SIInstrInfo::getNonSoftWaitcntOpcode(Opcode)));
----------------
Pierre-vh wrote:

> Note that SIInsertWaitcnts is iterative, so it may process the same block several times. If that happens then by the time the it runs the second time on a block, all remaining soft waitcnts will have been converted to hard ones by the first pass. Is that desirable for some reason?

So  my assumption is that it's a normal process. The first iteration will either eliminate the unneeded waitcnts, or make them non-soft. Further iterations on the block should act like they did before when there were no soft waitcnts.

I've renamed this function to also make it clearer because I think its current name was confusing. It's only called when we update a waitcnt (so when we know the waitcnt is needed) in order to "promote" the soft waitcnt into a normal one.

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