[PATCH] D49832: [SCEV] Don't expand Wrap predicate using inttoptr in ni addrspaces

[Sanjoy Das via Phabricator via llvm-commits]

sanjoy accepted this revision.
sanjoy added a comment.
This revision is now accepted and ready to land.

lgtm



================
Comment at: lib/Analysis/ScalarEvolutionExpander.cpp:2204
+  if (ARExpandTy->isPointerTy()) {
+    PointerType *ARPtrTy = cast<PointerType>(ARExpandTy);
+    const SCEV *MulS = SE.getSCEV(MulV);
----------------
This can be

```
if (auto *ARPtrTy = dyn_cast<PointerType>(ARExpandTy)) {
}
```



================
Comment at: lib/Analysis/ScalarEvolutionExpander.cpp:2208
+    Add = Builder.CreateBitCast(
+        expandAddToGEP(&StepArray[0], &StepArray[1], ARPtrTy, Ty, StartValue),
+        ARPtrTy);
----------------
This pattern is a bit confusing -- do you mind adding an overload for `expandAddToGEP` that takes one operand and does this internally?


================
Comment at: test/Analysis/LoopAccessAnalysis/wrapping-pointer-ni.ll:15
+
+define void @"japi1_align!_9477"(%jl_value_t addrspace(10)**) #0 {
+; LV-LAVEL: L26.lver.check
----------------
Do you mind cleaning up the test using instnamer?  Having `%1`, `%2` as instruction names make it difficult to change the test later since adding an instruction changes the numbering for all following instructions.

While you're at it, can you please also remove some of the unnecessary metadata here?


Repository:
  rL LLVM

https://reviews.llvm.org/D49832