[OpaquePtr][AArch64] Use elementtype on ldxr/stxr

Includes verifier changes checking the elementtype, clang codegen
changes to emit the elementtype, and ISel changes using the elementtype.

Reviewed By: #opaque-pointers, nikic

Differential Revision: https://reviews.llvm.org/D120527

4 files changed, 48 insertions, 42 deletions

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index a2699f5b3ea1..6383dfdd8950 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -9684,23 +9684,26 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
 
     QualType Ty = E->getType();
     llvm::Type *RealResTy = ConvertType(Ty);
-    llvm::Type *PtrTy = llvm::IntegerType::get(
-        getLLVMContext(), getContext().getTypeSize(Ty))->getPointerTo();
+    llvm::Type *IntTy =
+        llvm::IntegerType::get(getLLVMContext(), getContext().getTypeSize(Ty));
+    llvm::Type *PtrTy = IntTy->getPointerTo();
     LoadAddr = Builder.CreateBitCast(LoadAddr, PtrTy);
 
     Function *F = CGM.getIntrinsic(BuiltinID == AArch64::BI__builtin_arm_ldaex
                                        ? Intrinsic::aarch64_ldaxr
                                        : Intrinsic::aarch64_ldxr,
                                    PtrTy);
-    Value *Val = Builder.CreateCall(F, LoadAddr, "ldxr");
+    CallInst *Val = Builder.CreateCall(F, LoadAddr, "ldxr");
+    Val->addParamAttr(
+        0, Attribute::get(getLLVMContext(), Attribute::ElementType, IntTy));
 
     if (RealResTy->isPointerTy())
       return Builder.CreateIntToPtr(Val, RealResTy);
 
     llvm::Type *IntResTy = llvm::IntegerType::get(
         getLLVMContext(), CGM.getDataLayout().getTypeSizeInBits(RealResTy));
-    Val = Builder.CreateTruncOrBitCast(Val, IntResTy);
-    return Builder.CreateBitCast(Val, RealResTy);
+    return Builder.CreateBitCast(Builder.CreateTruncOrBitCast(Val, IntResTy),
+                                 RealResTy);
   }
 
   if ((BuiltinID == AArch64::BI__builtin_arm_strex ||
@@ -9748,7 +9751,10 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
                                        ? Intrinsic::aarch64_stlxr
                                        : Intrinsic::aarch64_stxr,
                                    StoreAddr->getType());
-    return Builder.CreateCall(F, {StoreVal, StoreAddr}, "stxr");
+    CallInst *CI = Builder.CreateCall(F, {StoreVal, StoreAddr}, "stxr");
+    CI->addParamAttr(
+        1, Attribute::get(getLLVMContext(), Attribute::ElementType, StoreTy));
+    return CI;
   }
 
   if (BuiltinID == AArch64::BI__getReg) {
diff --git a/clang/test/CodeGen/arm_acle.c b/clang/test/CodeGen/arm_acle.c
index 99c281633fc3..350aff581464 100644
--- a/clang/test/CodeGen/arm_acle.c
+++ b/clang/test/CodeGen/arm_acle.c
@@ -153,10 +153,10 @@ void test_dbg(void) {
 // AArch64-NEXT:    [[TMP0:%.*]] = bitcast i8* [[P:%.*]] to i32*
 // AArch64-NEXT:    br label [[DO_BODY_I:%.*]]
 // AArch64:       do.body.i:
-// AArch64-NEXT:    [[LDXR_I:%.*]] = call i64 @llvm.aarch64.ldxr.p0i32(i32* [[TMP0]]) [[ATTR3]]
+// AArch64-NEXT:    [[LDXR_I:%.*]] = call i64 @llvm.aarch64.ldxr.p0i32(i32* elementtype(i32) [[TMP0]]) [[ATTR3]]
 // AArch64-NEXT:    [[TMP1:%.*]] = trunc i64 [[LDXR_I]] to i32
 // AArch64-NEXT:    [[TMP2:%.*]] = zext i32 [[X:%.*]] to i64
-// AArch64-NEXT:    [[STXR_I:%.*]] = call i32 @llvm.aarch64.stxr.p0i32(i64 [[TMP2]], i32* [[TMP0]]) [[ATTR3]]
+// AArch64-NEXT:    [[STXR_I:%.*]] = call i32 @llvm.aarch64.stxr.p0i32(i64 [[TMP2]], i32* elementtype(i32) [[TMP0]]) [[ATTR3]]
 // AArch64-NEXT:    [[TOBOOL_I:%.*]] = icmp ne i32 [[STXR_I]], 0
 // AArch64-NEXT:    br i1 [[TOBOOL_I]], label [[DO_BODY_I]], label [[__SWP_EXIT:%.*]], [[LOOP6:!llvm.loop !.*]]
 // AArch64:       __swp.exit:
diff --git a/clang/test/CodeGen/builtins-arm-exclusive.c b/clang/test/CodeGen/builtins-arm-exclusive.c
index c6cf231659ec..5abe888e50ae 100644
--- a/clang/test/CodeGen/builtins-arm-exclusive.c
+++ b/clang/test/CodeGen/builtins-arm-exclusive.c
@@ -13,7 +13,7 @@ int test_ldrex(char *addr, long long *addr64, float *addrfloat) {
 // CHECK: [[INTRES:%.*]] = call i32 @llvm.arm.ldrex.p0i8(i8* %addr)
 // CHECK: trunc i32 [[INTRES]] to i8
 
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i8(i8* %addr)
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i8(i8* elementtype(i8) %addr)
 // CHECK-ARM64: trunc i64 [[INTRES]] to i8
 
   sum += __builtin_arm_ldrex((short *)addr);
@@ -22,7 +22,7 @@ int test_ldrex(char *addr, long long *addr64, float *addrfloat) {
 // CHECK: trunc i32 [[INTRES]] to i16
 
 // CHECK-ARM64: [[ADDR16:%.*]] = bitcast i8* %addr to i16*
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i16(i16* [[ADDR16]])
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i16(i16* elementtype(i16) [[ADDR16]])
 // CHECK-ARM64: trunc i64 [[INTRES]] to i16
 
   sum += __builtin_arm_ldrex((int *)addr);
@@ -30,7 +30,7 @@ int test_ldrex(char *addr, long long *addr64, float *addrfloat) {
 // CHECK: call i32 @llvm.arm.ldrex.p0i32(i32* [[ADDR32]])
 
 // CHECK-ARM64: [[ADDR32:%.*]] = bitcast i8* %addr to i32*
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i32(i32* [[ADDR32]])
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i32(i32* elementtype(i32) [[ADDR32]])
 // CHECK-ARM64: trunc i64 [[INTRES]] to i32
 
   sum += __builtin_arm_ldrex((long long *)addr);
@@ -39,13 +39,13 @@ int test_ldrex(char *addr, long long *addr64, float *addrfloat) {
 // CHECK: call { i32, i32 } @llvm.arm.ldrexd(i8* [[TMP5]])
 
 // CHECK-ARM64: [[ADDR64:%.*]] = bitcast i8* %addr to i64*
-// CHECK-ARM64: call i64 @llvm.aarch64.ldxr.p0i64(i64* [[ADDR64]])
+// CHECK-ARM64: call i64 @llvm.aarch64.ldxr.p0i64(i64* elementtype(i64) [[ADDR64]])
 
   sum += __builtin_arm_ldrex(addr64);
 // CHECK: [[ADDR64_AS8:%.*]] = bitcast i64* %addr64 to i8*
 // CHECK: call { i32, i32 } @llvm.arm.ldrexd(i8* [[ADDR64_AS8]])
 
-// CHECK-ARM64: call i64 @llvm.aarch64.ldxr.p0i64(i64* %addr64)
+// CHECK-ARM64: call i64 @llvm.aarch64.ldxr.p0i64(i64* elementtype(i64) %addr64)
 
   sum += __builtin_arm_ldrex(addrfloat);
 // CHECK: [[INTADDR:%.*]] = bitcast float* %addrfloat to i32*
@@ -53,7 +53,7 @@ int test_ldrex(char *addr, long long *addr64, float *addrfloat) {
 // CHECK: bitcast i32 [[INTRES]] to float
 
 // CHECK-ARM64: [[INTADDR:%.*]] = bitcast float* %addrfloat to i32*
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i32(i32* [[INTADDR]])
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i32(i32* elementtype(i32) [[INTADDR]])
 // CHECK-ARM64: [[TRUNCRES:%.*]] = trunc i64 [[INTRES]] to i32
 // CHECK-ARM64: bitcast i32 [[TRUNCRES]] to float
 
@@ -71,7 +71,7 @@ int test_ldrex(char *addr, long long *addr64, float *addrfloat) {
 
 // CHECK-ARM64: [[TMP4:%.*]] = bitcast i8* %addr to double*
 // CHECK-ARM64: [[TMP5:%.*]] = bitcast double* [[TMP4]] to i64*
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i64(i64* [[TMP5]])
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i64(i64* elementtype(i64) [[TMP5]])
 // CHECK-ARM64: bitcast i64 [[INTRES]] to double
 
   sum += *__builtin_arm_ldrex((int **)addr);
@@ -82,7 +82,7 @@ int test_ldrex(char *addr, long long *addr64, float *addrfloat) {
 
 // CHECK-ARM64: [[TMP4:%.*]] = bitcast i8* %addr to i32**
 // CHECK-ARM64: [[TMP5:%.*]] = bitcast i32** [[TMP4]] to i64*
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i64(i64* [[TMP5]])
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i64(i64* elementtype(i64) [[TMP5]])
 // CHECK-ARM64: inttoptr i64 [[INTRES]] to i32*
 
   sum += __builtin_arm_ldrex((struct Simple **)addr)->a;
@@ -93,7 +93,7 @@ int test_ldrex(char *addr, long long *addr64, float *addrfloat) {
 
 // CHECK-ARM64: [[TMP4:%.*]] = bitcast i8* %addr to %struct.Simple**
 // CHECK-ARM64: [[TMP5:%.*]] = bitcast %struct.Simple** [[TMP4]] to i64*
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i64(i64* [[TMP5]])
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldxr.p0i64(i64* elementtype(i64) [[TMP5]])
 // CHECK-ARM64: inttoptr i64 [[INTRES]] to %struct.Simple*
   return sum;
 }
@@ -106,7 +106,7 @@ int test_ldaex(char *addr, long long *addr64, float *addrfloat) {
 // CHECK: [[INTRES:%.*]] = call i32 @llvm.arm.ldaex.p0i8(i8* %addr)
 // CHECK: trunc i32 [[INTRES]] to i8
 
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i8(i8* %addr)
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i8(i8* elementtype(i8) %addr)
 // CHECK-ARM64: trunc i64 [[INTRES]] to i8
 
   sum += __builtin_arm_ldaex((short *)addr);
@@ -115,7 +115,7 @@ int test_ldaex(char *addr, long long *addr64, float *addrfloat) {
 // CHECK: trunc i32 [[INTRES]] to i16
 
 // CHECK-ARM64: [[ADDR16:%.*]] = bitcast i8* %addr to i16*
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i16(i16* [[ADDR16]])
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i16(i16* elementtype(i16) [[ADDR16]])
 // CHECK-ARM64: [[TRUNCRES:%.*]] = trunc i64 [[INTRES]] to i16
 
   sum += __builtin_arm_ldaex((int *)addr);
@@ -123,7 +123,7 @@ int test_ldaex(char *addr, long long *addr64, float *addrfloat) {
 // CHECK:  call i32 @llvm.arm.ldaex.p0i32(i32* [[ADDR32]])
 
 // CHECK-ARM64: [[ADDR32:%.*]] = bitcast i8* %addr to i32*
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i32(i32* [[ADDR32]])
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i32(i32* elementtype(i32) [[ADDR32]])
 // CHECK-ARM64: trunc i64 [[INTRES]] to i32
 
   sum += __builtin_arm_ldaex((long long *)addr);
@@ -132,13 +132,13 @@ int test_ldaex(char *addr, long long *addr64, float *addrfloat) {
 // CHECK: call { i32, i32 } @llvm.arm.ldaexd(i8* [[TMP5]])
 
 // CHECK-ARM64: [[ADDR64:%.*]] = bitcast i8* %addr to i64*
-// CHECK-ARM64: call i64 @llvm.aarch64.ldaxr.p0i64(i64* [[ADDR64]])
+// CHECK-ARM64: call i64 @llvm.aarch64.ldaxr.p0i64(i64* elementtype(i64) [[ADDR64]])
 
   sum += __builtin_arm_ldaex(addr64);
 // CHECK: [[ADDR64_AS8:%.*]] = bitcast i64* %addr64 to i8*
 // CHECK: call { i32, i32 } @llvm.arm.ldaexd(i8* [[ADDR64_AS8]])
 
-// CHECK-ARM64: call i64 @llvm.aarch64.ldaxr.p0i64(i64* %addr64)
+// CHECK-ARM64: call i64 @llvm.aarch64.ldaxr.p0i64(i64* elementtype(i64) %addr64)
 
   sum += __builtin_arm_ldaex(addrfloat);
 // CHECK: [[INTADDR:%.*]] = bitcast float* %addrfloat to i32*
@@ -146,7 +146,7 @@ int test_ldaex(char *addr, long long *addr64, float *addrfloat) {
 // CHECK: bitcast i32 [[INTRES]] to float
 
 // CHECK-ARM64: [[INTADDR:%.*]] = bitcast float* %addrfloat to i32*
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i32(i32* [[INTADDR]])
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i32(i32* elementtype(i32) [[INTADDR]])
 // CHECK-ARM64: [[TRUNCRES:%.*]] = trunc i64 [[INTRES]] to i32
 // CHECK-ARM64: bitcast i32 [[TRUNCRES]] to float
 
@@ -164,7 +164,7 @@ int test_ldaex(char *addr, long long *addr64, float *addrfloat) {
 
 // CHECK-ARM64: [[TMP4:%.*]] = bitcast i8* %addr to double*
 // CHECK-ARM64: [[TMP5:%.*]] = bitcast double* [[TMP4]] to i64*
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i64(i64* [[TMP5]])
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i64(i64* elementtype(i64) [[TMP5]])
 // CHECK-ARM64: bitcast i64 [[INTRES]] to double
 
   sum += *__builtin_arm_ldaex((int **)addr);
@@ -175,7 +175,7 @@ int test_ldaex(char *addr, long long *addr64, float *addrfloat) {
 
 // CHECK-ARM64: [[TMP4:%.*]] = bitcast i8* %addr to i32**
 // CHECK-ARM64: [[TMP5:%.*]] = bitcast i32** [[TMP4]] to i64*
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i64(i64* [[TMP5]])
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i64(i64* elementtype(i64) [[TMP5]])
 // CHECK-ARM64: inttoptr i64 [[INTRES]] to i32*
 
   sum += __builtin_arm_ldaex((struct Simple **)addr)->a;
@@ -186,7 +186,7 @@ int test_ldaex(char *addr, long long *addr64, float *addrfloat) {
 
 // CHECK-ARM64: [[TMP4:%.*]] = bitcast i8* %addr to %struct.Simple**
 // CHECK-ARM64: [[TMP5:%.*]] = bitcast %struct.Simple** [[TMP4]] to i64*
-// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i64(i64* [[TMP5]])
+// CHECK-ARM64: [[INTRES:%.*]] = call i64 @llvm.aarch64.ldaxr.p0i64(i64* elementtype(i64) [[TMP5]])
 // CHECK-ARM64: inttoptr i64 [[INTRES]] to %struct.Simple*
   return sum;
 }
@@ -199,21 +199,21 @@ int test_strex(char *addr) {
   res |= __builtin_arm_strex(4, addr);
 // CHECK: call i32 @llvm.arm.strex.p0i8(i32 4, i8* %addr)
 
-// CHECK-ARM64: call i32 @llvm.aarch64.stxr.p0i8(i64 4, i8* %addr)
+// CHECK-ARM64: call i32 @llvm.aarch64.stxr.p0i8(i64 4, i8* elementtype(i8) %addr)
 
   res |= __builtin_arm_strex(42, (short *)addr);
 // CHECK: [[ADDR16:%.*]] = bitcast i8* %addr to i16*
 // CHECK:  call i32 @llvm.arm.strex.p0i16(i32 42, i16* [[ADDR16]])
 
 // CHECK-ARM64: [[ADDR16:%.*]] = bitcast i8* %addr to i16*
-// CHECK-ARM64:  call i32 @llvm.aarch64.stxr.p0i16(i64 42, i16* [[ADDR16]])
+// CHECK-ARM64:  call i32 @llvm.aarch64.stxr.p0i16(i64 42, i16* elementtype(i16) [[ADDR16]])
 
   res |= __builtin_arm_strex(42, (int *)addr);
 // CHECK: [[ADDR32:%.*]] = bitcast i8* %addr to i32*
 // CHECK: call i32 @llvm.arm.strex.p0i32(i32 42, i32* [[ADDR32]])
 
 // CHECK-ARM64: [[ADDR32:%.*]] = bitcast i8* %addr to i32*
-// CHECK-ARM64: call i32 @llvm.aarch64.stxr.p0i32(i64 42, i32* [[ADDR32]])
+// CHECK-ARM64: call i32 @llvm.aarch64.stxr.p0i32(i64 42, i32* elementtype(i32) [[ADDR32]])
 
   res |= __builtin_arm_strex(42, (long long *)addr);
 // CHECK: store i64 42, i64* [[TMP:%.*]], align 8
@@ -226,7 +226,7 @@ int test_strex(char *addr) {
 // CHECK: call i32 @llvm.arm.strexd(i32 [[LO]], i32 [[HI]], i8* [[TMP5]])
 
 // CHECK-ARM64: [[ADDR64:%.*]] = bitcast i8* %addr to i64*
-// CHECK-ARM64: call i32 @llvm.aarch64.stxr.p0i64(i64 42, i64* [[ADDR64]])
+// CHECK-ARM64: call i32 @llvm.aarch64.stxr.p0i64(i64 42, i64* elementtype(i64) [[ADDR64]])
 
   res |= __builtin_arm_strex(2.71828f, (float *)addr);
 // CHECK: [[TMP4:%.*]] = bitcast i8* %addr to float*
@@ -235,7 +235,7 @@ int test_strex(char *addr) {
 
 // CHECK-ARM64: [[TMP4:%.*]] = bitcast i8* %addr to float*
 // CHECK-ARM64: [[TMP5:%.*]] = bitcast float* [[TMP4]] to i32*
-// CHECK-ARM64: call i32 @llvm.aarch64.stxr.p0i32(i64 1076754509, i32* [[TMP5]])
+// CHECK-ARM64: call i32 @llvm.aarch64.stxr.p0i32(i64 1076754509, i32* elementtype(i32) [[TMP5]])
 
   res |= __builtin_arm_strex(3.14159, (double *)addr);
 // CHECK: store double 3.141590e+00, double* [[TMP:%.*]], align 8
@@ -249,7 +249,7 @@ int test_strex(char *addr) {
 
 // CHECK-ARM64: [[TMP4:%.*]] = bitcast i8* %addr to double*
 // CHECK-ARM64: [[TMP5:%.*]] = bitcast double* [[TMP4]] to i64*
-// CHECK-ARM64: call i32 @llvm.aarch64.stxr.p0i64(i64 4614256650576692846, i64* [[TMP5]])
+// CHECK-ARM64: call i32 @llvm.aarch64.stxr.p0i64(i64 4614256650576692846, i64* elementtype(i64) [[TMP5]])
 
   res |= __builtin_arm_strex(&var, (struct Simple **)addr);
 // CHECK: [[TMP4:%.*]] = bitcast i8* %addr to %struct.Simple**
@@ -260,7 +260,7 @@ int test_strex(char *addr) {
 // CHECK-ARM64: [[TMP4:%.*]] = bitcast i8* %addr to %struct.Simple**
 // CHECK-ARM64: [[TMP5:%.*]] = bitcast %struct.Simple** [[TMP4]] to i64*
 // CHECK-ARM64: [[INTVAL:%.*]] = ptrtoint %struct.Simple* %var to i64
-// CHECK-ARM64: call i32 @llvm.aarch64.stxr.p0i64(i64 [[INTVAL]], i64* [[TMP5]])
+// CHECK-ARM64: call i32 @llvm.aarch64.stxr.p0i64(i64 [[INTVAL]], i64* elementtype(i64) [[TMP5]])
 
   return res;
 }
@@ -273,21 +273,21 @@ int test_stlex(char *addr) {
   res |= __builtin_arm_stlex(4, addr);
 // CHECK: call i32 @llvm.arm.stlex.p0i8(i32 4, i8* %addr)
 
-// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i8(i64 4, i8* %addr)
+// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i8(i64 4, i8* elementtype(i8) %addr)
 
   res |= __builtin_arm_stlex(42, (short *)addr);
 // CHECK: [[ADDR16:%.*]] = bitcast i8* %addr to i16*
 // CHECK:  call i32 @llvm.arm.stlex.p0i16(i32 42, i16* [[ADDR16]])
 
 // CHECK-ARM64: [[ADDR16:%.*]] = bitcast i8* %addr to i16*
-// CHECK-ARM64:  call i32 @llvm.aarch64.stlxr.p0i16(i64 42, i16* [[ADDR16]])
+// CHECK-ARM64:  call i32 @llvm.aarch64.stlxr.p0i16(i64 42, i16* elementtype(i16) [[ADDR16]])
 
   res |= __builtin_arm_stlex(42, (int *)addr);
 // CHECK: [[ADDR32:%.*]] = bitcast i8* %addr to i32*
 // CHECK: call i32 @llvm.arm.stlex.p0i32(i32 42, i32* [[ADDR32]])
 
 // CHECK-ARM64: [[ADDR32:%.*]] = bitcast i8* %addr to i32*
-// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i32(i64 42, i32* [[ADDR32]])
+// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i32(i64 42, i32* elementtype(i32) [[ADDR32]])
 
   res |= __builtin_arm_stlex(42, (long long *)addr);
 // CHECK: store i64 42, i64* [[TMP:%.*]], align 8
@@ -300,7 +300,7 @@ int test_stlex(char *addr) {
 // CHECK: call i32 @llvm.arm.stlexd(i32 [[LO]], i32 [[HI]], i8* [[TMP5]])
 
 // CHECK-ARM64: [[ADDR64:%.*]] = bitcast i8* %addr to i64*
-// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i64(i64 42, i64* [[ADDR64]])
+// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i64(i64 42, i64* elementtype(i64) [[ADDR64]])
 
   res |= __builtin_arm_stlex(2.71828f, (float *)addr);
 // CHECK: [[TMP4:%.*]] = bitcast i8* %addr to float*
@@ -309,7 +309,7 @@ int test_stlex(char *addr) {
 
 // CHECK-ARM64: [[TMP4:%.*]] = bitcast i8* %addr to float*
 // CHECK-ARM64: [[TMP5:%.*]] = bitcast float* [[TMP4]] to i32*
-// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i32(i64 1076754509, i32* [[TMP5]])
+// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i32(i64 1076754509, i32* elementtype(i32) [[TMP5]])
 
   res |= __builtin_arm_stlex(3.14159, (double *)addr);
 // CHECK: store double 3.141590e+00, double* [[TMP:%.*]], align 8
@@ -323,7 +323,7 @@ int test_stlex(char *addr) {
 
 // CHECK-ARM64: [[TMP4:%.*]] = bitcast i8* %addr to double*
 // CHECK-ARM64: [[TMP5:%.*]] = bitcast double* [[TMP4]] to i64*
-// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i64(i64 4614256650576692846, i64* [[TMP5]])
+// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i64(i64 4614256650576692846, i64* elementtype(i64) [[TMP5]])
 
   res |= __builtin_arm_stlex(&var, (struct Simple **)addr);
 // CHECK: [[TMP4:%.*]] = bitcast i8* %addr to %struct.Simple**
@@ -334,7 +334,7 @@ int test_stlex(char *addr) {
 // CHECK-ARM64: [[TMP4:%.*]] = bitcast i8* %addr to %struct.Simple**
 // CHECK-ARM64: [[TMP5:%.*]] = bitcast %struct.Simple** [[TMP4]] to i64*
 // CHECK-ARM64: [[INTVAL:%.*]] = ptrtoint %struct.Simple* %var to i64
-// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i64(i64 [[INTVAL]], i64* [[TMP5]])
+// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i64(i64 [[INTVAL]], i64* elementtype(i64) [[TMP5]])
 
   return res;
 }
diff --git a/clang/test/CodeGenCXX/builtins-arm-exclusive.cpp b/clang/test/CodeGenCXX/builtins-arm-exclusive.cpp
index 06f7a023adec..05cb330a4dda 100644
--- a/clang/test/CodeGenCXX/builtins-arm-exclusive.cpp
+++ b/clang/test/CodeGenCXX/builtins-arm-exclusive.cpp
@@ -7,7 +7,7 @@ bool b;
 // CHECK: call i32 @llvm.arm.ldrex.p0i8(i8* @b)
 
 // CHECK-ARM64-LABEL: @_Z10test_ldrexv()
-// CHECK-ARM64: call i64 @llvm.aarch64.ldxr.p0i8(i8* @b)
+// CHECK-ARM64: call i64 @llvm.aarch64.ldxr.p0i8(i8* elementtype(i8) @b)
 
 void test_ldrex() {
   b = __builtin_arm_ldrex(&b);
@@ -17,7 +17,7 @@ void test_ldrex() {
 // CHECK: %{{.*}} = call i32 @llvm.arm.strex.p0i8(i32 1, i8* @b)
 
 // CHECK-ARM64-LABEL: @_Z10tset_strexv()
-// CHECK-ARM64: %{{.*}} = call i32 @llvm.aarch64.stxr.p0i8(i64 1, i8* @b)
+// CHECK-ARM64: %{{.*}} = call i32 @llvm.aarch64.stxr.p0i8(i64 1, i8* elementtype(i8) @b)
 
 void tset_strex() {
   __builtin_arm_strex(true, &b);