Index: lib/Analysis/InstructionSimplify.cpp
===================================================================
--- lib/Analysis/InstructionSimplify.cpp
+++ lib/Analysis/InstructionSimplify.cpp
@@ -4467,10 +4467,12 @@
                                  const DominatorTree *DT, AssumptionCache *AC,
                                  OptimizationRemarkEmitter *ORE) {
   Value *Result;
+  if (auto *C = ConstantFoldInstruction(I, DL, TLI))
+    return C;
 
   switch (I->getOpcode()) {
   default:
-    Result = ConstantFoldInstruction(I, DL, TLI);
+    Result = nullptr;
     break;
   case Instruction::FAdd:
     Result = SimplifyFAddInst(I->getOperand(0), I->getOperand(1),
Index: test/Analysis/ConstantFolding/gep-constanfolding-error.ll
===================================================================
--- test/Analysis/ConstantFolding/gep-constanfolding-error.ll
+++ test/Analysis/ConstantFolding/gep-constanfolding-error.ll
@@ -44,7 +44,7 @@
   %9 = add i32 %f.promoted, %smax
   %10 = add i32 %9, 2
   call void @llvm.memset.p0i8.i32(i8* %scevgep, i8 %conv6, i32 %10, i32 1, i1 false)
-; CHECK:  call void @llvm.memset.p0i8.i32(i8* getelementptr inbounds ([6 x [6 x [7 x i8]]], [6 x [6 x [7 x i8]]]* @j, i32 0, i64 5, i64 4, i32 1), i8 %conv6, i32 1, i32 1, i1 false)
+; CHECK:  call void @llvm.memset.p0i8.i32(i8* getelementptr inbounds ([6 x [6 x [7 x i8]]], [6 x [6 x [7 x i8]]]* @j, i32 0, i{{32|64}} 5, i{{32|64}} 4, i32 1), i8 %conv6, i32 1, i32 1, i1 false)
 ; CHECK-NOT: call void @llvm.memset.p0i8.i32(i8* getelementptr ([6 x [6 x [7 x i8]]], [6 x [6 x [7 x i8]]]* @j, i64 1, i64 4, i64 4, i32 1)
   ret i32 0
 }
Index: test/Transforms/GVN/PRE/2011-06-01-NonLocalMemdepMiscompile.ll
===================================================================
--- test/Transforms/GVN/PRE/2011-06-01-NonLocalMemdepMiscompile.ll
+++ test/Transforms/GVN/PRE/2011-06-01-NonLocalMemdepMiscompile.ll
@@ -19,7 +19,7 @@
   br i1 undef, label %bb3, label %bb15
 
 ; CHECK: bb1:
-; CHECK: [[TMP:%.*]] = phi i8* [ getelementptr (i8, i8* null, i64 undef), %bb10 ], [ null, %bb ]
+; CHECK: [[TMP:%.*]] = phi i8* [ getelementptr inbounds (i8, i8* null, i64 undef), %bb10 ], [ null, %bb ]
 
 ; CHECK: bb1.bb15_crit_edge:
 ; CHECK: %tmp17.pre = load i8, i8* [[TMP]], align 1
Index: test/Transforms/InstSimplify/compare.ll
===================================================================
--- test/Transforms/InstSimplify/compare.ll
+++ test/Transforms/InstSimplify/compare.ll
@@ -69,7 +69,7 @@
 
 define i1 @PR31262() {
 ; CHECK-LABEL: @PR31262(
-; CHECK-NEXT:    ret i1 icmp uge (i32* getelementptr ([1 x i32], [1 x i32]* @a, i64 0, i64 undef), i32* getelementptr inbounds ([1 x i32], [1 x i32]* @a, i32 0, i32 0))
+; CHECK-NEXT:    ret i1 icmp uge (i32* getelementptr ([1 x i32], [1 x i32]* @a, i32 0, i32 undef), i32* getelementptr inbounds ([1 x i32], [1 x i32]* @a, i32 0, i32 0))
 ;
   %idx = getelementptr inbounds [1 x i32], [1 x i32]* @a, i64 0, i64 undef
   %cmp = icmp uge i32* %idx, getelementptr inbounds ([1 x i32], [1 x i32]* @a, i32 0, i32 0)
Index: test/Transforms/InstSimplify/simplify-nested-bitcast.ll
===================================================================
--- /dev/null
+++ test/Transforms/InstSimplify/simplify-nested-bitcast.ll
@@ -0,0 +1,54 @@
+; RUN: opt -always-inline -S %s | FileCheck %s
+%0 = type { i64, i64, i8 addrspace(1)*, i8 addrspace(1)* }
+%__aaa_struct = type { { i8**, i32, i32, i8*, %struct.__block_descriptor addrspace(1)* }, %0, [17 x i8], { i8**, i32, i32, i8*, %struct.__block_descriptor addrspace(1)* }, %0, [18 x i8] }
+%struct.__block_descriptor = type { i64, i64 }
+%struct.__block_literal_generic = type { i8*, i32, i32, i8*, %struct.__block_descriptor addrspace(1)* }
+
+@__aaa_struct_ptr = external addrspace(1) global %__aaa_struct
+@__aaa_const_init = constant %__aaa_struct { { i8**, i32, i32, i8*, %struct.__block_descriptor addrspace(1)* } { i8** null, i32 1342177280, i32 0, i8* bitcast (i32 (i8 addrspace(4)*, i32 addrspace(1)*)* @bl0_block_invoke to i8*), %struct.__block_descriptor addrspace(1)* bitcast (%0 addrspace(1)* getelementptr inbounds (%__aaa_struct, %__aaa_struct addrspace(1)* @__aaa_struct_ptr, i32 0, i32 1) to %struct.__block_descriptor addrspace(1)*) }, %0 { i64 0, i64 32, i8 addrspace(1)* getelementptr inbounds (%__aaa_struct, %__aaa_struct addrspace(1)* @__aaa_struct_ptr, i32 0, i32 2, i32 0), i8 addrspace(1)* null }, [17 x i8] c"bl0_block_invoke\00", { i8**, i32, i32, i8*, %struct.__block_descriptor addrspace(1)* } { i8** null, i32 1342177280, i32 0, i8* bitcast (i32 (i8 addrspace(4)*, i32 addrspace(1)*)* @__f1_block_invoke to i8*), %struct.__block_descriptor addrspace(1)* bitcast (%0 addrspace(1)* getelementptr inbounds (%__aaa_struct, %__aaa_struct addrspace(1)* @__aaa_struct_ptr, i32 0, i32 4) to %struct.__block_descriptor addrspace(1)*) }, %0 { i64 0, i64 32, i8 addrspace(1)* getelementptr inbounds (%__aaa_struct, %__aaa_struct addrspace(1)* @__aaa_struct_ptr, i32 0, i32 5, i32 0), i8 addrspace(1)* null }, [18 x i8] c"__f1_block_invoke\00" }
+
+; Function Attrs: alwaysinline norecurse nounwind readonly
+define i32 @bl0_block_invoke(i8 addrspace(4)* nocapture readnone, i32 addrspace(1)* nocapture readonly) #0 {
+entry:
+  %2 = load i32, i32 addrspace(1)* %1, align 4
+  %mul = shl nsw i32 %2, 1
+  ret i32 %mul
+}
+
+; Function Attrs: alwaysinline nounwind
+define i32 @f0(i32 addrspace(1)*, i32 (i32 addrspace(1)*) addrspace(4)*) #1 {
+entry:
+  %block.literal = bitcast i32 (i32 addrspace(1)*) addrspace(4)* %1 to %struct.__block_literal_generic addrspace(4)*
+  %2 = getelementptr inbounds %struct.__block_literal_generic, %struct.__block_literal_generic addrspace(4)* %block.literal, i64 0, i32 3
+  %3 = bitcast i32 (i32 addrspace(1)*) addrspace(4)* %1 to i8 addrspace(4)*
+  %4 = bitcast i8* addrspace(4)* %2 to i32 (i8 addrspace(4)*, i32 addrspace(1)*)* addrspace(4)*
+  %5 = load i32 (i8 addrspace(4)*, i32 addrspace(1)*)*, i32 (i8 addrspace(4)*, i32 addrspace(1)*)* addrspace(4)* %4, align 8
+  %call = tail call i32 %5(i8 addrspace(4)* %3, i32 addrspace(1)* %0) #2
+  ret i32 %call
+}
+
+; CHECK-LABEL: define void @f1
+; CHECK: %1 = load i32 (i8 addrspace(4)*, i32 addrspace(1)*)*, i32 (i8 addrspace(4)*, i32 addrspace(1)*)* addrspace(4)* bitcast (i8* addrspace(4)* getelementptr inbounds (%__aaa_struct, %__aaa_struct addrspace(4)* addrspacecast (%__aaa_struct addrspace(1)* @__aaa_struct_ptr to %__aaa_struct addrspace(4)*), i64 0, i32 0, i32 3) to i32 (i8 addrspace(4)*, i32 addrspace(1)*)* addrspace(4)*), align 8
+
+; Function Attrs: alwaysinline nounwind
+define void @f1(i32 addrspace(1)*) #1 {
+entry:
+  %call = tail call i32 @f0(i32 addrspace(1)* %0, i32 (i32 addrspace(1)*) addrspace(4)* addrspacecast (i32 (i32 addrspace(1)*) addrspace(1)* bitcast (%__aaa_struct addrspace(1)* @__aaa_struct_ptr to i32 (i32 addrspace(1)*) addrspace(1)*) to i32 (i32 addrspace(1)*) addrspace(4)*)) #3
+  store i32 %call, i32 addrspace(1)* %0, align 4
+  %call1 = tail call i32 @f0(i32 addrspace(1)* %0, i32 (i32 addrspace(1)*) addrspace(4)* addrspacecast (i32 (i32 addrspace(1)*) addrspace(1)* bitcast ({ i8**, i32, i32, i8*, %struct.__block_descriptor addrspace(1)* } addrspace(1)* getelementptr inbounds (%__aaa_struct, %__aaa_struct addrspace(1)* @__aaa_struct_ptr, i32 0, i32 3) to i32 (i32 addrspace(1)*) addrspace(1)*) to i32 (i32 addrspace(1)*) addrspace(4)*)) #3
+  store i32 %call1, i32 addrspace(1)* %0, align 4
+  ret void
+}
+
+; Function Attrs: alwaysinline norecurse nounwind readonly
+define i32 @__f1_block_invoke(i8 addrspace(4)* nocapture readnone, i32 addrspace(1)* nocapture readonly) #0 {
+entry:
+  %2 = load i32, i32 addrspace(1)* %1, align 4
+  %add = add nsw i32 %2, 1
+  ret i32 %add
+}
+
+attributes #0 = { alwaysinline norecurse nounwind readonly }
+attributes #1 = { alwaysinline nounwind }
+attributes #2 = { nobuiltin nounwind }
+attributes #3 = { nobuiltin }
Index: test/Transforms/InstSimplify/vector_gep.ll
===================================================================
--- test/Transforms/InstSimplify/vector_gep.ll
+++ test/Transforms/InstSimplify/vector_gep.ll
@@ -51,7 +51,7 @@
   ret <4 x i8*> %gep
 
 ; CHECK-LABEL: @test5
-; CHECK-NEXT: ret <4 x i8*> getelementptr (i8, <4 x i8*> <i8* inttoptr (i64 1 to i8*), i8* inttoptr (i64 2 to i8*), i8* inttoptr (i64 3 to i8*), i8* inttoptr (i64 4 to i8*)>, <4 x i32> <i32 1, i32 1, i32 1, i32 1>)
+; CHECK-NEXT: ret <4 x i8*> getelementptr (i8, <4 x i8*> <i8* inttoptr (i64 1 to i8*), i8* inttoptr (i64 2 to i8*), i8* inttoptr (i64 3 to i8*), i8* inttoptr (i64 4 to i8*)>, <4 x i64> <i64 1, i64 1, i64 1, i64 1>)
 }
 
 @v = global [24 x [42 x [3 x i32]]] zeroinitializer, align 16
@@ -61,4 +61,4 @@
 ; CHECK-NEXT: ret <16 x i32*> getelementptr ([24 x [42 x [3 x i32]]], [24 x [42 x [3 x i32]]]* @v, <16 x i64> zeroinitializer, <16 x i64> zeroinitializer, <16 x i64> <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7, i64 8, i64 9, i64 10, i64 11, i64 12, i64 13, i64 14, i64 15>, <16 x i64> zeroinitializer)
   %VectorGep = getelementptr [24 x [42 x [3 x i32]]], [24 x [42 x [3 x i32]]]* @v, i64 0, i64 0, <16 x i64> <i64 0, i64 1, i64 2, i64 3, i64 4, i64 5, i64 6, i64 7, i64 8, i64 9, i64 10, i64 11, i64 12, i64 13, i64 14, i64 15>, i64 0
   ret <16 x i32*> %VectorGep
-}
\ No newline at end of file
+}