[llvm] [AMDGPU] Use alias info to relax waitcounts for LDS DMA (PR #74537)

[Stanislav Mekhanoshin via llvm-commits]

================
@@ -703,8 +713,37 @@ void WaitcntBrackets::updateByEvent(const SIInstrInfo *TII,
         setRegScore(RegNo, T, CurrScore);
       }
     }
-    if (Inst.mayStore() && (TII->isDS(Inst) || mayWriteLDSThroughDMA(Inst))) {
-      setRegScore(SQ_MAX_PGM_VGPRS + EXTRA_VGPR_LDS, T, CurrScore);
+    if (Inst.mayStore() &&
+        (TII->isDS(Inst) || TII->mayWriteLDSThroughDMA(Inst))) {
+      // MUBUF and FLAT LDS DMA operations need a wait on vmcnt before LDS
+      // written can be accessed. A load from LDS to VMEM does not need a wait.
+      unsigned Slot = 0;
+      for (const auto *MemOp : Inst.memoperands()) {
+        if (!MemOp->isStore() ||
+            MemOp->getAddrSpace() != AMDGPUAS::LOCAL_ADDRESS)
+          continue;
+        // Comparing just AA info does not guarantee memoperands are equal
----------------
rampitec wrote:

Speaking of Value, this is how a real program's MIR look like. These are all loads and stores of the same arrays. You may see that Value there is not helpful, only alias scope is (look at the second memory operand, LDS store):

```
  BUFFER_LOAD_DWORD_LDS_OFFEN %51:vgpr_32, %48:sgpr_128, %52:sreg_32, 0, 0, 0, implicit $exec, implicit $m0 :: (dereferenceable load (s32) from `ptr addrspace(1) poison`, align 1, !alias.scope !7, !noalias !10, addrspace 1), (dereferenceable store (s32) into %ir.10, align 1, !alias.scope !7, !noalias !10, addrspace 3)
  BUFFER_LOAD_DWORD_LDS_OFFEN killed %54:vgpr_32, %48:sgpr_128, %52:sreg_32, 0, 0, 0, implicit $exec, implicit $m0 :: (dereferenceable load (s32) from `ptr addrspace(1) poison`, align 1, !alias.scope !7, !noalias !10, addrspace 1), (dereferenceable store (s32) into %ir.12, align 1, !alias.scope !7, !noalias !10, addrspace 3)
  BUFFER_LOAD_DWORD_LDS_OFFEN killed %56:vgpr_32, %48:sgpr_128, %52:sreg_32, 0, 0, 0, implicit $exec, implicit $m0 :: (dereferenceable load (s32) from `ptr addrspace(1) poison`, align 1, !alias.scope !7, !noalias !10, addrspace 1), (dereferenceable store (s32) into %ir.14, align 1, !alias.scope !7, !noalias !10, addrspace 3)
  BUFFER_LOAD_DWORD_LDS_OFFEN killed %58:vgpr_32, %48:sgpr_128, %52:sreg_32, 0, 0, 0, implicit $exec, implicit $m0 :: (dereferenceable load (s32) from `ptr addrspace(1) poison`, align 1, !alias.scope !7, !noalias !10, addrspace 1), (dereferenceable store (s32) into %ir.16, align 1, !alias.scope !7, !noalias !10, addrspace 3)
```

https://github.com/llvm/llvm-project/pull/74537