Index: llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
===================================================================
--- llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
+++ llvm/lib/Transforms/InstCombine/InstructionCombining.cpp
@@ -2433,13 +2433,21 @@
     // analysis of unions. If "A" is also a bitcast, wait for A/X to be merged.
     unsigned OffsetBits = DL.getIndexTypeSizeInBits(GEPType);
     APInt Offset(OffsetBits, 0);
-    if (!isa<BitCastInst>(SrcOp) && GEP.accumulateConstantOffset(DL, Offset)) {
+
+    // If the bitcast argument is an allocation, The bitcast is for convertion to
+    // actual type of allocation. Removing such bitcasts, results in having GEPs
+    // with i8* base and pure byte offsets. That means GEP is not aware of Struct
+    // or array hierarchy. 
+    // By avoiding such GEPs, phi translation and MemoryDependencyAnalysis have 
+    // a better chance to succeed
+    if (!isa<BitCastInst>(SrcOp) && GEP.accumulateConstantOffset(DL, Offset) &&
+        !isAllocationFn(SrcOp, &TLI)) {
       // If this GEP instruction doesn't move the pointer, just replace the GEP
       // with a bitcast of the real input to the dest type.
       if (!Offset) {
         // If the bitcast is of an allocation, and the allocation will be
         // converted to match the type of the cast, don't touch this.
-        if (isa<AllocaInst>(SrcOp) || isAllocationFn(SrcOp, &TLI)) {
+        if (isa<AllocaInst>(SrcOp)) {
           // See if the bitcast simplifies, if so, don't nuke this GEP yet.
           if (Instruction *I = visitBitCast(*BCI)) {
             if (I != BCI) {
Index: llvm/test/Transforms/InstCombine/getelementptr.ll
===================================================================
--- llvm/test/Transforms/InstCombine/getelementptr.ll
+++ llvm/test/Transforms/InstCombine/getelementptr.ll
@@ -1273,4 +1273,23 @@
   ret i8* %gep
 }
 
+declare noalias i8* @malloc(i64) nounwind
+
+define i32 @test_gep_bitcast_malloc(%struct.A* %a) {
+; CHECK-LABEL: @test_gep_bitcast_malloc(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[CALL:%.*]] = call noalias dereferenceable_or_null(16) i8* @malloc(i64 16)
+; CHECK-NEXT:    [[B:%.*]] = bitcast i8* [[CALL]] to %struct.A*
+; CHECK-NEXT:    [[G3:%.*]] = getelementptr [[STRUCT_A:%.*]], %struct.A* [[B]], i64 0, i32 2
+; CHECK-NEXT:    [[A_C:%.*]] = load i32, i32* [[G3]], align 4
+; CHECK-NEXT:    ret i32 [[A_C]]
+;
+entry:
+  %call = call noalias i8* @malloc(i64 16) #2
+  %B = bitcast i8* %call to %struct.A*
+  %g3 = getelementptr %struct.A, %struct.A* %B, i32 0, i32 2
+  %a_c = load i32, i32* %g3, align 4
+  ret i32 %a_c
+}
+
 !0 = !{!"branch_weights", i32 2, i32 10}