Fixing an infinite loop problem in InstCombine

[mfawaz]

Summary

This issues started happening after
https://github.com/llvm/llvm-project/commit/b373b5990d5991a920c421b21a352e4ccf4c4993
Basically, if the memcpy is volatile, the collectUsers() function should
return false, just like we do for volatile loads.

[jeroen.dobbelaere]

LGTM. Thanks !

This change should also go on the 13.x branch.

llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp

private:
   SmallSetVector<Instruction *, 4> Worklist;
   MapVector<Value *, Value *> WorkMap;
   InstCombinerImpl &IC;
 };
 } // end anonymous namespace
 
 bool PointerReplacer::collectUsers(Instruction &I) {
   for (auto U : I.users()) {
-    Instruction *Inst = cast<Instruction>(&*U);
-    if (LoadInst *Load = dyn_cast<LoadInst>(Inst)) {
+    auto *Inst = cast<Instruction>(&*U);
+    if (auto *Load = dyn_cast<LoadInst>(Inst)) {
       if (Load->isVolatile())
         return false;
       Worklist.insert(Load);
     } else if (isa<GetElementPtrInst>(Inst) || isa<BitCastInst>(Inst)) {
       Worklist.insert(Inst);
       if (!collectUsers(*Inst))
         return false;
-    } else if (isa<MemTransferInst>(Inst)) {
+    } else if (auto *MI = dyn_cast<MemTransferInst>(Inst)) {
+      if (MI->isVolatile())
+        return false;
       Worklist.insert(Inst);
     } else if (Inst->isLifetimeStartOrEnd()) {
       continue;
     } else {
       LLVM_DEBUG(dbgs() << "Cannot handle pointer user: " << *U << '\n');
       return false;
     }
   }

llvm/test/Transforms/InstCombine/memcpy-from-global.ll

entry:
   %b = bitcast [4 x float]* %a to i8*
   call void @llvm.lifetime.start.p0i8(i64 16, i8* %b)
   call void @llvm.memcpy.p0i8.p1i8.i64(i8* align 4 %b, i8 addrspace(1)* align 4 bitcast ([4 x float] addrspace(1)* @I to i8 addrspace(1)*), i64 16, i1 false)
   %g = getelementptr inbounds [4 x float], [4 x float]* %a, i64 0, i64 %i
   %r = load float, float* %g, align 4
   ret float %r
 }
 
+; If the memcpy is volatile, it should not be removed
+define float @test11_volatile(i64 %i) {
+; CHECK-LABEL: @test11_volatile(
+; CHECK-NEXT: entry:
+; CHECK-NEXT: alloca
+; CHECK: call void @llvm.lifetime.start.p0i8
+; CHECK: call void @llvm.memcpy
+
+entry:
+  %a = alloca [4 x float], align 4
+  %b = bitcast [4 x float]* %a to i8*
+  call void @llvm.lifetime.start.p0i8(i64 16, i8* %b)
+  call void @llvm.memcpy.p0i8.p1i8.i64(i8* align 4 %b, i8 addrspace(1)* align 4 bitcast ([4 x float] addrspace(1)* @I to i8 addrspace(1)*), i64 16, i1 true)
+  %g = getelementptr inbounds [4 x float], [4 x float]* %a, i64 0, i64 %i
+  %r = load float, float* %g, align 4
+  ret float %r
+}
+
 attributes #0 = { null_pointer_is_valid }