[ConstantFolding] Don't create illegal (non-integral) inttoptrs

[majnemer]

LGTM with nits

lib/Analysis/ConstantFolding.cpp

831

auto *

test/Transforms/InstCombine/non-integral-pointers.ll

1

Can this be in -instsimplify?

[sanjoy]

test/Transforms/InstCombine/non-integral-pointers.ll

1

-instsimplify for GEPs calls SimplifyGEPInst which does not dispatch into ConstantFoldInstruction.

lib/Analysis/ConstantFolding.cpp

Constant *SymbolicallyEvaluateGEP(const GEPOperator *GEP,
   APInt BasePtr(BitWidth, 0);
   if (auto *CE = dyn_cast<ConstantExpr>(Ptr)) {
     if (CE->getOpcode() == Instruction::IntToPtr) {
       if (auto *Base = dyn_cast<ConstantInt>(CE->getOperand(0)))
         BasePtr = Base->getValue().zextOrTrunc(BitWidth);
     }
   }
 
-  if (Ptr->isNullValue() || BasePtr != 0) {
+  PointerType *PTy = cast<PointerType>(Ptr->getType());
+  if ((Ptr->isNullValue() || BasePtr != 0) &&
+      !DL.isNonIntegralPointerType(PTy)) {
     Constant *C = ConstantInt::get(Ptr->getContext(), Offset + BasePtr);
     return ConstantExpr::getIntToPtr(C, ResTy);
   }
 
   // Otherwise form a regular getelementptr. Recompute the indices so that
   // we eliminate over-indexing of the notional static type array bounds.
   // This makes it easy to determine if the getelementptr is "inbounds".
   // Also, this helps GlobalOpt do SROA on GlobalVariables.
-  Type *Ty = Ptr->getType();
-  assert(Ty->isPointerTy() && "Forming regular GEP of non-pointer type");
+  Type *Ty = PTy;
   SmallVector<Constant *, 32> NewIdxs;
 
   do {
     if (!Ty->isStructTy()) {
       if (Ty->isPointerTy()) {
         // The only pointer indexing we'll do is on the first index of the GEP.
         if (!NewIdxs.empty())
           break;

test/Transforms/InstCombine/non-integral-pointers.ll

+; RUN: opt -instcombine -S < %s | FileCheck %s
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128-ni:4"
+target triple = "x86_64-unknown-linux-gnu"
+
+define i8 addrspace(4)* @f_0() {
+; CHECK-LABEL: @f_0(
+; CHECK: ret i8 addrspace(4)* getelementptr (i8, i8 addrspace(4)* null, i64 50)
+  %result = getelementptr i8, i8 addrspace(4)* null, i64 50
+  ret i8 addrspace(4)* %result
+}
+
+define i8 addrspace(3)* @f_1() {
+; inttoptr is fine here since addrspace(3) is integral.
+
+; CHECK-LABEL: @f_1(
+; CHECK: ret i8 addrspace(3)* inttoptr (i64 50 to i8 addrspace(3)*)
+  %result = getelementptr i8, i8 addrspace(3)* null, i64 50
+  ret i8 addrspace(3)* %result
+}