[CGP] Support opaque pointers in address mode fold

Rather than inspecting the pointer element type, use the access
type of the load/store/atomicrmw/cmpxchg.

In the process of doing this, simplify the logic by storing the
address + type in MemoryUses, rather than an Instruction + Operand
pair (which was then used to fetch the address).

Author: nikic

llvm/lib/CodeGen/CodeGenPrepare.cpp

@@ -4859,10 +4859,9 @@ static constexpr int MaxMemoryUsesToScan = 20;
 
 /// Recursively walk all the uses of I until we find a memory use.
 /// If we find an obviously non-foldable instruction, return true.
-/// Add the ultimately found memory instructions to MemoryUses.
+/// Add accessed addresses and types to MemoryUses.
 static bool FindAllMemoryUses(
-    Instruction *I,
-    SmallVectorImpl<std::pair<Instruction *, unsigned>> &MemoryUses,
+    Instruction *I, SmallVectorImpl<std::pair<Value *, Type *>> &MemoryUses,
     SmallPtrSetImpl<Instruction *> &ConsideredInsts, const TargetLowering &TLI,
     const TargetRegisterInfo &TRI, bool OptSize, ProfileSummaryInfo *PSI,
     BlockFrequencyInfo *BFI, int SeenInsts = 0) {
@@ -4883,31 +4882,28 @@ static bool FindAllMemoryUses(
 
     Instruction *UserI = cast<Instruction>(U.getUser());
     if (LoadInst *LI = dyn_cast<LoadInst>(UserI)) {
-      MemoryUses.push_back(std::make_pair(LI, U.getOperandNo()));
+      MemoryUses.push_back({U.get(), LI->getType()});
       continue;
     }
 
     if (StoreInst *SI = dyn_cast<StoreInst>(UserI)) {
-      unsigned opNo = U.getOperandNo();
-      if (opNo != StoreInst::getPointerOperandIndex())
+      if (U.getOperandNo() != StoreInst::getPointerOperandIndex())
         return true; // Storing addr, not into addr.
-      MemoryUses.push_back(std::make_pair(SI, opNo));
+      MemoryUses.push_back({U.get(), SI->getValueOperand()->getType()});
       continue;
     }
 
     if (AtomicRMWInst *RMW = dyn_cast<AtomicRMWInst>(UserI)) {
-      unsigned opNo = U.getOperandNo();
-      if (opNo != AtomicRMWInst::getPointerOperandIndex())
+      if (U.getOperandNo() != AtomicRMWInst::getPointerOperandIndex())
         return true; // Storing addr, not into addr.
-      MemoryUses.push_back(std::make_pair(RMW, opNo));
+      MemoryUses.push_back({U.get(), RMW->getValOperand()->getType()});
       continue;
     }
 
     if (AtomicCmpXchgInst *CmpX = dyn_cast<AtomicCmpXchgInst>(UserI)) {
-      unsigned opNo = U.getOperandNo();
-      if (opNo != AtomicCmpXchgInst::getPointerOperandIndex())
+      if (U.getOperandNo() != AtomicCmpXchgInst::getPointerOperandIndex())
         return true; // Storing addr, not into addr.
-      MemoryUses.push_back(std::make_pair(CmpX, opNo));
+      MemoryUses.push_back({U.get(), CmpX->getCompareOperand()->getType()});
       continue;
     }
 
@@ -5017,7 +5013,7 @@ isProfitableToFoldIntoAddressingMode(Instruction *I, ExtAddrMode &AMBefore,
   // we can remove the addressing mode and effectively trade one live register
   // for another (at worst.)  In this context, folding an addressing mode into
   // the use is just a particularly nice way of sinking it.
-  SmallVector<std::pair<Instruction*,unsigned>, 16> MemoryUses;
+  SmallVector<std::pair<Value *, Type *>, 16> MemoryUses;
   SmallPtrSet<Instruction*, 16> ConsideredInsts;
   if (FindAllMemoryUses(I, MemoryUses, ConsideredInsts, TLI, TRI, OptSize,
                         PSI, BFI))
@@ -5033,18 +5029,10 @@ isProfitableToFoldIntoAddressingMode(Instruction *I, ExtAddrMode &AMBefore,
   // growth since most architectures have some reasonable small and fast way to
   // compute an effective address.  (i.e LEA on x86)
   SmallVector<Instruction*, 32> MatchedAddrModeInsts;
-  for (unsigned i = 0, e = MemoryUses.size(); i != e; ++i) {
-    Instruction *User = MemoryUses[i].first;
-    unsigned OpNo = MemoryUses[i].second;
-
-    // Get the access type of this use.  If the use isn't a pointer, we don't
-    // know what it accesses.
-    Value *Address = User->getOperand(OpNo);
-    PointerType *AddrTy = dyn_cast<PointerType>(Address->getType());
-    if (!AddrTy)
-      return false;
-    Type *AddressAccessTy = AddrTy->getElementType();
-    unsigned AS = AddrTy->getAddressSpace();
+  for (const std::pair<Value *, Type *> &Pair : MemoryUses) {
+    Value *Address = Pair.first;
+    Type *AddressAccessTy = Pair.second;
+    unsigned AS = Address->getType()->getPointerAddressSpace();
 
     // Do a match against the root of this address, ignoring profitability. This
     // will tell us if the addressing mode for the memory operation will

llvm/test/CodeGen/X86/merge_store.ll

@@ -1,5 +1,6 @@
 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -opaque-pointers | FileCheck %s
 
 define void @merge_store(i32* nocapture %a) {
 ; CHECK-LABEL: merge_store: