[llvm] [AMDGPU][PromoteAlloca] Whole-function alloca promotion to vector (PR #84735)

[Matt Arsenault via llvm-commits]

================
@@ -225,6 +249,55 @@ FunctionPass *llvm::createAMDGPUPromoteAllocaToVector() {
   return new AMDGPUPromoteAllocaToVector();
 }
 
+static void collectAllocaUses(AllocaInst &Alloca,
+                              SmallVectorImpl<Use *> &Uses) {
+  SmallVector<Instruction *, 4> WorkList({&Alloca});
+  while (!WorkList.empty()) {
+    auto *Cur = WorkList.pop_back_val();
+    for (auto &U : Cur->uses()) {
+      Uses.push_back(&U);
+
+      if (isa<GetElementPtrInst>(U.getUser()) || isa<BitCastInst>(U.getUser()))
+        WorkList.push_back(cast<Instruction>(U.getUser()));
+    }
+  }
+}
+
+void AMDGPUPromoteAllocaImpl::sortAllocasToPromote(
+    SmallVectorImpl<AllocaInst *> &Allocas) {
+  DenseMap<AllocaInst *, unsigned> Scores;
+
+  LLVM_DEBUG(dbgs() << "Before sorting allocas:\n"; for (auto *A
+                                                         : Allocas) dbgs()
+                                                    << "  " << *A << "\n";);
+
+  for (auto *Alloca : Allocas) {
+    LLVM_DEBUG(dbgs() << "Scoring: " << *Alloca << "\n");
+    unsigned &Score = Scores[Alloca];
+    // Increment score by one for each user + a bonus for users within loops.
+    SmallVector<Use *, 8> Uses;
+    collectAllocaUses(*Alloca, Uses);
+    for (auto *U : Uses) {
+      Instruction *Inst = cast<Instruction>(U->getUser());
+      if (isa<GetElementPtrInst>(Inst) || isa<BitCastInst>(Inst))
+        continue;
+      unsigned UserScore =
+          1 + (LoopUserWeight * LI.getLoopDepth(Inst->getParent()));
+      LLVM_DEBUG(dbgs() << "  [+" << UserScore << "]:\t" << *Inst << "\n");
+      Score += UserScore;
+    }
+    LLVM_DEBUG(dbgs() << "  => Final Score:" << Score << "\n");
+  }
+
+  stable_sort(Allocas, [&](AllocaInst *A, AllocaInst *B) {
+    return Scores.at(A) > Scores.at(B);
+  });
+
+  LLVM_DEBUG(dbgs() << "After sorting allocas:\n"; for (auto *A
+                                                        : Allocas) dbgs()
+                                                   << "  " << *A << "\n";);
----------------
arsenm wrote:

```suggestion
  LLVM_DEBUG(
    dbgs() << "After sorting allocas:\n"; 
    for (auto *A : Allocas) 
      dbgs() << "  " << *A << '\n';
  );
```

https://github.com/llvm/llvm-project/pull/84735