[InstCombine] Restrict a GEP transform to avoid changing provenance

This is an alternative to D98120. Herein, instead of deleting the transformation entirely, we check
that the underlying objects are both the same and therefore this transformation wouldn't incur a
provenance change, if applied.

https://alive2.llvm.org/ce/z/SYF_yv

Reviewed By: lebedev.ri

Differential Revision: https://reviews.llvm.org/D98588

Author: nagisa

llvm/lib/Transforms/InstCombine/InstructionCombining.cpp

@@ -2173,14 +2173,15 @@ Instruction *InstCombinerImpl::visitGetElementPtrInst(GetElementPtrInst &GEP) {
           Matched = true;
       }
 
-      if (Matched) {
-        // Canonicalize (gep i8* X, (ptrtoint Y)-(ptrtoint X))
-        // to (bitcast Y)
-        Value *Y;
-        if (match(V, m_Sub(m_PtrToInt(m_Value(Y)),
-                           m_PtrToInt(m_Specific(GEP.getOperand(0))))))
-          return CastInst::CreatePointerBitCastOrAddrSpaceCast(Y, GEPType);
-      }
+      // Canonicalize (gep i8* X, (ptrtoint Y)-(ptrtoint X)) to (bitcast Y), but
+      // only if both point to the same underlying object (otherwise provenance
+      // is not necessarily retained).
+      Value *Y;
+      Value *X = GEP.getOperand(0);
+      if (Matched &&
+          match(V, m_Sub(m_PtrToInt(m_Value(Y)), m_PtrToInt(m_Specific(X)))) &&
+          getUnderlyingObject(X) == getUnderlyingObject(Y))
+        return CastInst::CreatePointerBitCastOrAddrSpaceCast(Y, GEPType);
     }
   }
 

llvm/test/Transforms/InstCombine/getelementptr.ll

@@ -1087,7 +1087,11 @@ define %struct.C* @test44i(%struct.C* %c1, %struct.C* %c2) {
 
 define %struct.C* @test45(%struct.C* %c1, %struct.C** %c2) {
 ; CHECK-LABEL: @test45(
-; CHECK-NEXT:    [[GEP:%.*]] = bitcast %struct.C** [[C2:%.*]] to %struct.C*
+; CHECK-NEXT:    [[PTRTOINT1:%.*]] = ptrtoint %struct.C* [[C1:%.*]] to i64
+; CHECK-NEXT:    [[PTRTOINT2:%.*]] = ptrtoint %struct.C** [[C2:%.*]] to i64
+; CHECK-NEXT:    [[SUB:%.*]] = sub i64 [[PTRTOINT2]], [[PTRTOINT1]]
+; CHECK-NEXT:    [[SHR:%.*]] = sdiv i64 [[SUB]], 7
+; CHECK-NEXT:    [[GEP:%.*]] = getelementptr inbounds [[STRUCT_C:%.*]], %struct.C* [[C1]], i64 [[SHR]]
 ; CHECK-NEXT:    ret %struct.C* [[GEP]]
 ;
   %ptrtoint1 = ptrtoint %struct.C* %c1 to i64