[clang] [lld] [llvm] [IR] Change representation of getelementptr inrange (PR #84341)

[Nikita Popov via cfe-commits]

================
@@ -47,28 +47,66 @@ static bool splitGlobal(GlobalVariable &GV) {
   if (!Init)
     return false;
 
-  // Verify that each user of the global is an inrange getelementptr constant.
-  // From this it follows that any loads from or stores to that global must use
-  // a pointer derived from an inrange getelementptr constant, which is
-  // sufficient to allow us to apply the splitting transform.
+  const DataLayout &DL = GV.getParent()->getDataLayout();
+  const StructLayout *SL = DL.getStructLayout(Init->getType());
+  ArrayRef<TypeSize> MemberOffsets = SL->getMemberOffsets();
+  unsigned IndexWidth = DL.getIndexTypeSizeInBits(GV.getType());
+
+  // Verify that each user of the global is an inrange getelementptr constant,
+  // and collect information on how it relates to the global.
+  struct GEPInfo {
+    GEPOperator *GEP;
+    unsigned MemberIndex;
+    APInt MemberRelativeOffset;
+
+    GEPInfo(GEPOperator *GEP, unsigned MemberIndex, APInt MemberRelativeOffset)
+        : GEP(GEP), MemberIndex(MemberIndex),
+          MemberRelativeOffset(std::move(MemberRelativeOffset)) {}
+  };
+  SmallVector<GEPInfo> Infos;
   for (User *U : GV.users()) {
-    if (!isa<Constant>(U))
+    auto *GEP = dyn_cast<GEPOperator>(U);
+    if (!GEP)
       return false;
 
-    auto *GEP = dyn_cast<GEPOperator>(U);
-    if (!GEP || !GEP->getInRangeIndex() || *GEP->getInRangeIndex() != 1 ||
-        !isa<ConstantInt>(GEP->getOperand(1)) ||
-        !cast<ConstantInt>(GEP->getOperand(1))->isZero() ||
-        !isa<ConstantInt>(GEP->getOperand(2)))
+    std::optional<ConstantRange> InRange = GEP->getInRange();
+    if (!InRange)
+      return false;
+
+    APInt Offset(IndexWidth, 0);
+    if (!GEP->accumulateConstantOffset(DL, Offset))
+      return false;
+
+    // Determine source-relative inrange.
+    ConstantRange SrcInRange = InRange->sextOrTrunc(IndexWidth).add(Offset);
+
+    // Check that the GEP offset is in the range (treating upper bound as
+    // inclusive here).
+    if (!SrcInRange.contains(Offset) && SrcInRange.getUpper() != Offset)
+      return false;
+
+    // Find which struct member the range corresponds to.
+    if (SrcInRange.getLower().uge(SL->getSizeInBytes()))
       return false;
+
+    unsigned MemberIndex =
+        SL->getElementContainingOffset(SrcInRange.getLower().getZExtValue());
----------------
nikic wrote:

I think this is already implied by the check directly before this (note that it uses uge rather than sge, so will also reject negative numbers -- unless the struct size becomes negative, of course...).

https://github.com/llvm/llvm-project/pull/84341