[llvm] [AMDGPU][NFC] Use a worklist and remember results (PR #129175)

[Johannes Doerfert via llvm-commits]

================
@@ -257,26 +259,43 @@ class AMDGPUInformationCache : public InformationCache {
   }
 
   /// Get the constant access bitmap for \p C.
-  uint8_t getConstantAccess(const Constant *C,
-                            SmallPtrSetImpl<const Constant *> &Visited) {
-    auto It = ConstantStatus.find(C);
+  uint8_t getConstantAccess(const Constant *C) {
+    const auto &It = ConstantStatus.find(C);
     if (It != ConstantStatus.end())
-      return It->second;
+      return It->second.value();
 
-    uint8_t Result = 0;
-    if (isDSAddress(C))
-      Result = DS_GLOBAL;
+    SmallPtrSet<const Constant *, 8> Visited;
+    SmallVector<const Constant *> Worklist;
+    Worklist.push_back(C);
+    Visited.insert(C);
 
-    if (const auto *CE = dyn_cast<ConstantExpr>(C))
-      Result |= visitConstExpr(CE);
+    uint8_t Result = 0;
+    while (Result != CS_WORST && !Worklist.empty()) {
+      const Constant *CurC = Worklist.pop_back_val();
 
-    for (const Use &U : C->operands()) {
-      const auto *OpC = dyn_cast<Constant>(U);
-      if (!OpC || !Visited.insert(OpC).second)
+      std::optional<uint8_t> &CurCResultOrNone = ConstantStatus[CurC];
----------------
jdoerfert wrote:

We can insert C early, sure, but the operands (CurC) will invalidate the iterator later.
So, if C is contained, we are done, if not, we insert one value at a time (with a single operator[] lookup), and then insert C at the end. Does that make sense?

https://github.com/llvm/llvm-project/pull/129175