[PATCH] Minor refactoring of GEP handling in isDereferenceablePointer

[Sanjoy Das]

I'm not sure that this is an NFC change -- I think the previous code will return `false` on the `%1` below while the new code will return `true`.

  %struct.A = type { [8 x i8], [5 x i8] }
  
  define i8 @f(%struct.A* %a) {
    %1 = getelementptr inbounds %struct.A, %struct.A* %a, i64 0, i32 0, i64 10
    %2 = load i8, i8* %1
    ret i8 %2
  }

Please add a test case that shows this difference.


================
Comment at: lib/Analysis/ValueTracking.cpp:2987
@@ +2986,3 @@
+    uint64_t LoadSize = DL.getTypeStoreSize(Ty);
+    Type* BaseType = Base->getType()->getPointerElementType();
+    if ((Offset + LoadSize).ugt(DL.getTypeAllocSize(BaseType)))
----------------
LLVM style is `Type *`.

================
Comment at: lib/Analysis/ValueTracking.cpp:2989
@@ +2988,3 @@
+    if ((Offset + LoadSize).ugt(DL.getTypeAllocSize(BaseType)))
+      return false;
+    
----------------
Can we simplify this to `return (Offset + LoadSize).ule(DL.getTypeAllocSize(BaseType))`?

http://reviews.llvm.org/D9874

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/