[llvm] [AMDGPU] Remove s_delay_alu for VALU->SGPR->SALU (PR #127212)

[Ana Mihajlovic via llvm-commits]

================
@@ -254,12 +255,15 @@ class AMDGPUInsertDelayAlu : public MachineFunctionPass {
       }
     }
 
-    void advanceByNum(DelayType Type, unsigned Cycles, unsigned VALUNum) {
+    void advanceByNum(DelayType Type, unsigned Cycles, unsigned SGPRWriteVALUNum) {
       iterator Next;
       for (auto I = begin(), E = end(); I != E; I = Next) {
         Next = std::next(I);
-        if (I->second.VALUNum >= VALUNum && I->second.advance(Type, Cycles))
+        if (I->second.VALUNum >= SGPRWriteVALUNum && I->second.VALUCycles > 0){
----------------
mihajlovicana wrote:

advance increments VALUNum and we don't want to do that here because it will result in bad VALU_DEP in some cases

https://github.com/llvm/llvm-project/pull/127212