[NFC] Remove redundant loads when has_device_addr is used.

[jyu2]

Summary

It is caused by regenerate captured var value when processing the
has_device_addr, the captured var value has been generated in
GenerateOpenMPCapturedVars and passed as Arg in generateInfoForCapture.
The fix just use Arg instead regenerate, just same as is_device_ptr.

[ABataev]

Whyt it is NFC? I assume it fixes the dereferencing of references, right?

[jyu2]

In D137313#3904805, @ABataev wrote:

Whyt it is NFC? I assume it fixes the dereferencing of references, right?

Thanks Alexey.

I really don't expect run time change for this change. The change only change BP/PP of the map instead I was using

CGF.EmitLValue(E).getPointer(CGF);

or

CGF.EmitScalarExpr(E);

to generate BP/PP.

[ABataev]

Do we have a runtime test for this? Would be good to try to test it if the offloading fails and the host version is executed instead.

[jyu2]

In D137313#3906487, @ABataev wrote:

Do we have a runtime test for this? Would be good to try to test it if the offloading fails and the host version is executed instead.

I have runtime test in https://reviews.llvm.org/D134268
openmp/libomptarget/test/mapping/target_has_device_addr.c where

void bar() {

short x[10];
short *xp = &x[0];

x[1] = 111;

#pragma omp target data map(tofrom : xp [0:2]) use_device_addr(xp [0:2])
#pragma omp target has_device_addr(xp [0:2])

{
  xp[1] = 222;
  // CHECK: 222
  printf("%d %p\n", xp[1], &xp[1]);
}
// CHECK: 222
printf("%d %p\n", xp[1], &xp[1]);

}

[ABataev]

In D137313#3906527, @jyu2 wrote:

In D137313#3906487, @ABataev wrote:

Do we have a runtime test for this? Would be good to try to test it if the offloading fails and the host version is executed instead.

I have runtime test in https://reviews.llvm.org/D134268
openmp/libomptarget/test/mapping/target_has_device_addr.c where

void bar() {

short x[10];
short *xp = &x[0];

x[1] = 111;

#pragma omp target data map(tofrom : xp [0:2]) use_device_addr(xp [0:2])
#pragma omp target has_device_addr(xp [0:2])

{
  xp[1] = 222;
  // CHECK: 222
  printf("%d %p\n", xp[1], &xp[1]);
}
// CHECK: 222
printf("%d %p\n", xp[1], &xp[1]);

}

Need to add a reference

[jyu2]

Thanks Alexey. New runtime test is added.

[ABataev]

LG

[tianshilei1992]

nit: this doesn't look like NFC.

[jyu2]

de14befa7730093ff3d46c7628fa1084f251e98d

[ronlieb]

[AMD Official Use Only - General]

Please look into AMDGPU bot fail
https://lab.llvm.org/buildbot/#/builders/193/builds/21319

Get Outlook for iOShttps://aka.ms/o0ukef

---

From: Jennifer Yu via Phabricator <[e-mail]>
Sent: Friday, November 4, 2022 5:40:19 PM
To: [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>
Cc: [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; Pandey, Amit <[e-mail]>; Islam, Saiyedul <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; Lieberman, Ron <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; Balasubrmanian, Vignesh <[e-mail]>; [e-mail] <[e-mail]>; Kumar N, Bhuvanendra <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; Tomar, Sourabh Singh <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; [e-mail] <[e-mail]>; Song, Ruiling <[e-mail]>
Subject: [PATCH] D137313: [NFC] Remove redundant loads when has_device_addr is used.

Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.

jyu2 closed this revision.
jyu2 added a comment.

de14befa7730093ff3d46c7628fa1084f251e98d https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Freviews.llvm.org%2FrGde14befa7730093ff3d46c7628fa1084f251e98d&amp;data=05%7C01%7Cron.lieberman%40amd.com%7C89e7e812ba444904677f08dabeb58de0%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C638031984301236587%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&amp;sdata=91XJgV2YZU9jqHYiXbrS4d5t95xt2Bro%2FwEMG6SpUY0%3D&amp;reserved=0

Repository:

rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION

https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Freviews.llvm.org%2FD137313%2Fnew%2F&data=05%7C01%7Cron.lieberman%40amd.com%7C89e7e812ba444904677f08dabeb58de0%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C638031984301236587%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=o9eGJWg86IGPzt2G4SlgX6VTxpoWRlAEhb2TBO%2BcGTw%3D&reserved=0

https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Freviews.llvm.org%2FD137313&amp;data=05%7C01%7Cron.lieberman%40amd.com%7C89e7e812ba444904677f08dabeb58de0%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0%7C638031984301236587%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&amp;sdata=VPD4ootaA99%2FEMcRn34ONBJTzQYWvfTqqh2ErRLrSaA%3D&amp;reserved=0

clang/lib/CodeGen/CGOpenMPRuntime.cpp

void generateInfoForCapture(const CapturedStmt::Capture *Cap,
     // for map(to: lambda): skip here, processing it in
     // generateDefaultMapInfo
     if (LambdasMap.count(VD))
       return;
 
     // If this declaration appears in a is_device_ptr clause we just have to
     // pass the pointer by value. If it is a reference to a declaration, we just
     // pass its value.
-    if (VD && DevPointersMap.count(VD)) {
+    if (VD && (DevPointersMap.count(VD) || HasDevAddrsMap.count(VD))) {
       CombinedInfo.Exprs.push_back(VD);
       CombinedInfo.BasePointers.emplace_back(Arg, VD);
       CombinedInfo.Pointers.push_back(Arg);
       CombinedInfo.Sizes.push_back(CGF.Builder.CreateIntCast(
           CGF.getTypeSize(CGF.getContext().VoidPtrTy), CGF.Int64Ty,
           /*isSigned=*/true));
       CombinedInfo.Types.push_back(
           (Cap->capturesVariable() ? OMP_MAP_TO : OMP_MAP_LITERAL) |
           OMP_MAP_TARGET_PARAM);
       CombinedInfo.Mappers.push_back(nullptr);
       return;
     }
-    if (VD && HasDevAddrsMap.count(VD)) {
-      auto I = HasDevAddrsMap.find(VD);
-      CombinedInfo.Exprs.push_back(VD);
-      Expr *E = nullptr;
-      for (auto &MCL : I->second) {
-        E = MCL.begin()->getAssociatedExpression();
-        break;
-      }
-      llvm::Value *Ptr = nullptr;
-      if (E->isGLValue())
-        Ptr = CGF.EmitLValue(E).getPointer(CGF);
-      else
-        Ptr = CGF.EmitScalarExpr(E);
-      CombinedInfo.BasePointers.emplace_back(Ptr, VD);
-      CombinedInfo.Pointers.push_back(Ptr);
-      CombinedInfo.Sizes.push_back(CGF.Builder.CreateIntCast(
-          CGF.getTypeSize(CGF.getContext().VoidPtrTy), CGF.Int64Ty,
-          /*isSigned=*/true));
-      CombinedInfo.Types.push_back(
-          (Cap->capturesVariable() ? OMP_MAP_TO : OMP_MAP_LITERAL) |
-          OMP_MAP_TARGET_PARAM);
-      CombinedInfo.Mappers.push_back(nullptr);
-      return;
-    }
 
     using MapData =
         std::tuple<OMPClauseMappableExprCommon::MappableExprComponentListRef,
                    OpenMPMapClauseKind, ArrayRef<OpenMPMapModifierKind>, bool,
                    const ValueDecl *, const Expr *>;
     SmallVector<MapData, 4> DeclComponentLists;
     // For member fields list in is_device_ptr, store it in
     // DeclComponentLists for generating components info.

clang/test/OpenMP/target_has_device_addr_codegen.cpp

 // CHECK-NEXT:    br i1 [[TMP16]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 // CHECK:       omp_offload.failed:
 // CHECK-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l145(ptr [[K]]) #[[ATTR5:[0-9]+]]
 // CHECK-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 // CHECK:       omp_offload.cont:
 // CHECK-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[Z]], align 8
 // CHECK-NEXT:    store ptr [[TMP17]], ptr [[TMP]], align 8
 // CHECK-NEXT:    [[TMP18:%.*]] = load ptr, ptr [[TMP]], align 8
-// CHECK-NEXT:    [[TMP19:%.*]] = load ptr, ptr [[TMP]], align 8
 // CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-NEXT:    store ptr [[TMP19]], ptr [[TMP20]], align 8
+// CHECK-NEXT:    store ptr [[TMP18]], ptr [[TMP20]], align 8
 // CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK-NEXT:    store ptr [[TMP19]], ptr [[TMP21]], align 8
+// CHECK-NEXT:    store ptr [[TMP18]], ptr [[TMP21]], align 8
 // CHECK-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
 // CHECK-NEXT:    store ptr null, ptr [[TMP22]], align 8
 // CHECK-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
 // CHECK-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
 // CHECK-NEXT:    [[KERNEL_ARGS4:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
 // CHECK-NEXT:    store i32 1, ptr [[TMP25]], align 4
 // CHECK-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1

 // CHECK-NEXT:    br i1 [[TMP51]], label [[OMP_OFFLOAD_FAILED11:%.*]], label [[OMP_OFFLOAD_CONT12:%.*]]
 // CHECK:       omp_offload.failed11:
 // CHECK-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l149(ptr [[AA]]) #[[ATTR5]]
 // CHECK-NEXT:    br label [[OMP_OFFLOAD_CONT12]]
 // CHECK:       omp_offload.cont12:
 // CHECK-NEXT:    [[TMP52:%.*]] = load ptr, ptr [[RAA]], align 8
 // CHECK-NEXT:    store ptr [[TMP52]], ptr [[_TMP13]], align 8
 // CHECK-NEXT:    [[TMP53:%.*]] = load ptr, ptr [[_TMP13]], align 8
-// CHECK-NEXT:    [[TMP54:%.*]] = load ptr, ptr [[_TMP13]], align 8
 // CHECK-NEXT:    [[TMP55:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
-// CHECK-NEXT:    store ptr [[TMP54]], ptr [[TMP55]], align 8
+// CHECK-NEXT:    store ptr [[TMP53]], ptr [[TMP55]], align 8
 // CHECK-NEXT:    [[TMP56:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
-// CHECK-NEXT:    store ptr [[TMP54]], ptr [[TMP56]], align 8
+// CHECK-NEXT:    store ptr [[TMP53]], ptr [[TMP56]], align 8
 // CHECK-NEXT:    [[TMP57:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS16]], i64 0, i64 0
 // CHECK-NEXT:    store ptr null, ptr [[TMP57]], align 8
 // CHECK-NEXT:    [[TMP58:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS14]], i32 0, i32 0
 // CHECK-NEXT:    [[TMP59:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS15]], i32 0, i32 0
 // CHECK-NEXT:    [[KERNEL_ARGS17:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK-NEXT:    [[TMP60:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 0
 // CHECK-NEXT:    store i32 1, ptr [[TMP60]], align 4
 // CHECK-NEXT:    [[TMP61:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS17]], i32 0, i32 1

 // CHECK-NEXT:    store i64 0, ptr [[TMP84]], align 8
 // CHECK-NEXT:    [[TMP85:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153.region_id, ptr [[KERNEL_ARGS23]])
 // CHECK-NEXT:    [[TMP86:%.*]] = icmp ne i32 [[TMP85]], 0
 // CHECK-NEXT:    br i1 [[TMP86]], label [[OMP_OFFLOAD_FAILED24:%.*]], label [[OMP_OFFLOAD_CONT25:%.*]]
 // CHECK:       omp_offload.failed24:
 // CHECK-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l153(ptr [[H]]) #[[ATTR5]]
 // CHECK-NEXT:    br label [[OMP_OFFLOAD_CONT25]]
 // CHECK:       omp_offload.cont25:
-// CHECK-NEXT:    [[ARRAYDECAY:%.*]] = getelementptr inbounds [5 x i32], ptr [[DA]], i64 0, i64 0
 // CHECK-NEXT:    [[TMP87:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS26]], i32 0, i32 0
-// CHECK-NEXT:    store ptr [[ARRAYDECAY]], ptr [[TMP87]], align 8
+// CHECK-NEXT:    store ptr [[DA]], ptr [[TMP87]], align 8
 // CHECK-NEXT:    [[TMP88:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS27]], i32 0, i32 0
-// CHECK-NEXT:    store ptr [[ARRAYDECAY]], ptr [[TMP88]], align 8
+// CHECK-NEXT:    store ptr [[DA]], ptr [[TMP88]], align 8
 // CHECK-NEXT:    [[TMP89:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS28]], i64 0, i64 0
 // CHECK-NEXT:    store ptr null, ptr [[TMP89]], align 8
 // CHECK-NEXT:    [[TMP90:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS26]], i32 0, i32 0
 // CHECK-NEXT:    [[TMP91:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS27]], i32 0, i32 0
 // CHECK-NEXT:    [[KERNEL_ARGS29:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK-NEXT:    [[TMP92:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 0
 // CHECK-NEXT:    store i32 1, ptr [[TMP92]], align 4
 // CHECK-NEXT:    [[TMP93:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS29]], i32 0, i32 1

 // CHECK-NEXT:    br i1 [[TMP16]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 // CHECK:       omp_offload.failed:
 // CHECK-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIiET_S0__l123(ptr [[K]]) #[[ATTR5]]
 // CHECK-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 // CHECK:       omp_offload.cont:
 // CHECK-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[Z]], align 8
 // CHECK-NEXT:    store ptr [[TMP17]], ptr [[TMP]], align 8
 // CHECK-NEXT:    [[TMP18:%.*]] = load ptr, ptr [[TMP]], align 8
-// CHECK-NEXT:    [[TMP19:%.*]] = load ptr, ptr [[TMP]], align 8
 // CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-NEXT:    store ptr [[TMP19]], ptr [[TMP20]], align 8
+// CHECK-NEXT:    store ptr [[TMP18]], ptr [[TMP20]], align 8
 // CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK-NEXT:    store ptr [[TMP19]], ptr [[TMP21]], align 8
+// CHECK-NEXT:    store ptr [[TMP18]], ptr [[TMP21]], align 8
 // CHECK-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
 // CHECK-NEXT:    store ptr null, ptr [[TMP22]], align 8
 // CHECK-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
 // CHECK-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
 // CHECK-NEXT:    [[KERNEL_ARGS4:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
 // CHECK-NEXT:    store i32 1, ptr [[TMP25]], align 4
 // CHECK-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1

 // CHECK-NEXT:    br i1 [[TMP16]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 // CHECK:       omp_offload.failed:
 // CHECK-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z5tmainIPiET_S1__l123(ptr [[K]]) #[[ATTR5]]
 // CHECK-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 // CHECK:       omp_offload.cont:
 // CHECK-NEXT:    [[TMP17:%.*]] = load ptr, ptr [[Z]], align 8
 // CHECK-NEXT:    store ptr [[TMP17]], ptr [[TMP]], align 8
 // CHECK-NEXT:    [[TMP18:%.*]] = load ptr, ptr [[TMP]], align 8
-// CHECK-NEXT:    [[TMP19:%.*]] = load ptr, ptr [[TMP]], align 8
 // CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
-// CHECK-NEXT:    store ptr [[TMP19]], ptr [[TMP20]], align 8
+// CHECK-NEXT:    store ptr [[TMP18]], ptr [[TMP20]], align 8
 // CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
-// CHECK-NEXT:    store ptr [[TMP19]], ptr [[TMP21]], align 8
+// CHECK-NEXT:    store ptr [[TMP18]], ptr [[TMP21]], align 8
 // CHECK-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS3]], i64 0, i64 0
 // CHECK-NEXT:    store ptr null, ptr [[TMP22]], align 8
 // CHECK-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS1]], i32 0, i32 0
 // CHECK-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS2]], i32 0, i32 0
 // CHECK-NEXT:    [[KERNEL_ARGS4:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
 // CHECK-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 0
 // CHECK-NEXT:    store i32 1, ptr [[TMP25]], align 4
 // CHECK-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS4]], i32 0, i32 1

clang/test/OpenMP/target_has_device_addr_codegen_01.cpp

 // CHECK-NEXT:    [[VLA:%.*]] = alloca float, i64 [[TMP1]], align 4
 // CHECK-NEXT:    store i64 [[TMP1]], ptr [[__VLA_EXPR0]], align 8
 // CHECK-NEXT:    call void @_ZN1SC1Ev(ptr noundef nonnull align 8 dereferenceable(40) [[S]])
 // CHECK-NEXT:    call void @_ZN1S3fooEv(ptr noundef nonnull align 8 dereferenceable(40) [[S]])
 // CHECK-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[REF]], align 8
 // CHECK-NEXT:    store ptr [[TMP3]], ptr [[TMP]], align 8
 // CHECK-NEXT:    [[TMP4:%.*]] = load ptr, ptr [[PTR]], align 8
 // CHECK-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[TMP]], align 8
-// CHECK-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[PTR]], align 8
-// CHECK-NEXT:    [[TMP7:%.*]] = load ptr, ptr [[TMP]], align 8
-// CHECK-NEXT:    [[ARRAYDECAY:%.*]] = getelementptr inbounds [4 x float], ptr [[ARR]], i64 0, i64 0
-// CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-NEXT:    store ptr [[A]], ptr [[TMP8]], align 8
-// CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-NEXT:    store ptr [[A]], ptr [[TMP10]], align 8
-// CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK-NEXT:    store ptr null, ptr [[TMP12]], align 8
-// CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK-NEXT:    store ptr [[TMP6]], ptr [[TMP13]], align 8
-// CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK-NEXT:    store ptr [[TMP6]], ptr [[TMP15]], align 8
-// CHECK-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
+// CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-NEXT:    store ptr [[A]], ptr [[TMP6]], align 8
+// CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-NEXT:    store ptr [[A]], ptr [[TMP7]], align 8
+// CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK-NEXT:    store ptr null, ptr [[TMP8]], align 8
+// CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK-NEXT:    store ptr [[TMP4]], ptr [[TMP9]], align 8
+// CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
+// CHECK-NEXT:    store ptr [[TMP4]], ptr [[TMP10]], align 8
+// CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
+// CHECK-NEXT:    store ptr null, ptr [[TMP11]], align 8
+// CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK-NEXT:    store ptr [[TMP5]], ptr [[TMP12]], align 8
+// CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK-NEXT:    store ptr [[TMP5]], ptr [[TMP13]], align 8
+// CHECK-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
+// CHECK-NEXT:    store ptr null, ptr [[TMP14]], align 8
+// CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 3
+// CHECK-NEXT:    store ptr [[ARR]], ptr [[TMP15]], align 8
+// CHECK-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 3
+// CHECK-NEXT:    store ptr [[ARR]], ptr [[TMP16]], align 8
+// CHECK-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 3
 // CHECK-NEXT:    store ptr null, ptr [[TMP17]], align 8
-// CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK-NEXT:    store ptr [[TMP7]], ptr [[TMP18]], align 8
-// CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK-NEXT:    store ptr [[TMP7]], ptr [[TMP20]], align 8
-// CHECK-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
-// CHECK-NEXT:    store ptr null, ptr [[TMP22]], align 8
-// CHECK-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 3
-// CHECK-NEXT:    store ptr [[ARRAYDECAY]], ptr [[TMP23]], align 8
-// CHECK-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 3
-// CHECK-NEXT:    store ptr [[ARRAYDECAY]], ptr [[TMP25]], align 8
-// CHECK-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 3
-// CHECK-NEXT:    store ptr null, ptr [[TMP27]], align 8
-// CHECK-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 4
-// CHECK-NEXT:    store i64 [[TMP1]], ptr [[TMP28]], align 8
-// CHECK-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 4
-// CHECK-NEXT:    store i64 [[TMP1]], ptr [[TMP30]], align 8
-// CHECK-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 4
-// CHECK-NEXT:    store ptr null, ptr [[TMP32]], align 8
-// CHECK-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 5
-// CHECK-NEXT:    store ptr [[VLA]], ptr [[TMP33]], align 8
-// CHECK-NEXT:    [[TMP35:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 5
-// CHECK-NEXT:    store ptr [[VLA]], ptr [[TMP35]], align 8
-// CHECK-NEXT:    [[TMP37:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 5
-// CHECK-NEXT:    store ptr null, ptr [[TMP37]], align 8
-// CHECK-NEXT:    [[TMP38:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-NEXT:    [[TMP39:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 4
+// CHECK-NEXT:    store i64 [[TMP1]], ptr [[TMP18]], align 8
+// CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 4
+// CHECK-NEXT:    store i64 [[TMP1]], ptr [[TMP19]], align 8
+// CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 4
+// CHECK-NEXT:    store ptr null, ptr [[TMP20]], align 8
+// CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 5
+// CHECK-NEXT:    store ptr [[VLA]], ptr [[TMP21]], align 8
+// CHECK-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 5
+// CHECK-NEXT:    store ptr [[VLA]], ptr [[TMP22]], align 8
+// CHECK-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 5
+// CHECK-NEXT:    store ptr null, ptr [[TMP23]], align 8
+// CHECK-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [6 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
 // CHECK-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK-NEXT:    [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-NEXT:    store i32 1, ptr [[TMP40]], align 4
-// CHECK-NEXT:    [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK-NEXT:    store i32 6, ptr [[TMP41]], align 4
-// CHECK-NEXT:    [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK-NEXT:    store ptr [[TMP38]], ptr [[TMP42]], align 8
-// CHECK-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK-NEXT:    store ptr [[TMP39]], ptr [[TMP43]], align 8
-// CHECK-NEXT:    [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK-NEXT:    store ptr @.offload_sizes, ptr [[TMP44]], align 8
-// CHECK-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK-NEXT:    store ptr @.offload_maptypes, ptr [[TMP45]], align 8
-// CHECK-NEXT:    [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK-NEXT:    store ptr null, ptr [[TMP46]], align 8
-// CHECK-NEXT:    [[TMP47:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK-NEXT:    store ptr null, ptr [[TMP47]], align 8
-// CHECK-NEXT:    [[TMP48:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK-NEXT:    store i64 0, ptr [[TMP48]], align 8
-// CHECK-NEXT:    [[TMP49:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27.region_id, ptr [[KERNEL_ARGS]])
-// CHECK-NEXT:    [[TMP50:%.*]] = icmp ne i32 [[TMP49]], 0
-// CHECK-NEXT:    br i1 [[TMP50]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK-NEXT:    store i32 1, ptr [[TMP26]], align 4
+// CHECK-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK-NEXT:    store i32 6, ptr [[TMP27]], align 4
+// CHECK-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK-NEXT:    store ptr [[TMP24]], ptr [[TMP28]], align 8
+// CHECK-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK-NEXT:    store ptr [[TMP25]], ptr [[TMP29]], align 8
+// CHECK-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK-NEXT:    store ptr @.offload_sizes, ptr [[TMP30]], align 8
+// CHECK-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK-NEXT:    store ptr @.offload_maptypes, ptr [[TMP31]], align 8
+// CHECK-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK-NEXT:    store ptr null, ptr [[TMP32]], align 8
+// CHECK-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK-NEXT:    store ptr null, ptr [[TMP33]], align 8
+// CHECK-NEXT:    [[TMP34:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK-NEXT:    store i64 0, ptr [[TMP34]], align 8
+// CHECK-NEXT:    [[TMP35:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1:[0-9]+]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27.region_id, ptr [[KERNEL_ARGS]])
+// CHECK-NEXT:    [[TMP36:%.*]] = icmp ne i32 [[TMP35]], 0
+// CHECK-NEXT:    br i1 [[TMP36]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 // CHECK:       omp_offload.failed:
 // CHECK-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27(ptr [[A]], ptr [[TMP4]], ptr [[TMP5]], ptr [[ARR]], i64 [[TMP1]], ptr [[VLA]]) #[[ATTR5:[0-9]+]]
 // CHECK-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 // CHECK:       omp_offload.cont:
-// CHECK-NEXT:    [[TMP51:%.*]] = load float, ptr [[A]], align 4
-// CHECK-NEXT:    [[CONV1:%.*]] = fptosi float [[TMP51]] to i32
+// CHECK-NEXT:    [[TMP37:%.*]] = load float, ptr [[A]], align 4
+// CHECK-NEXT:    [[CONV1:%.*]] = fptosi float [[TMP37]] to i32
 // CHECK-NEXT:    store i32 [[CONV1]], ptr [[RETVAL]], align 4
-// CHECK-NEXT:    [[TMP52:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
-// CHECK-NEXT:    call void @llvm.stackrestore(ptr [[TMP52]])
-// CHECK-NEXT:    [[TMP53:%.*]] = load i32, ptr [[RETVAL]], align 4
-// CHECK-NEXT:    ret i32 [[TMP53]]
+// CHECK-NEXT:    [[TMP38:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
+// CHECK-NEXT:    call void @llvm.stackrestore(ptr [[TMP38]])
+// CHECK-NEXT:    [[TMP39:%.*]] = load i32, ptr [[RETVAL]], align 4
+// CHECK-NEXT:    ret i32 [[TMP39]]
 //
 //
 // CHECK-LABEL: define {{[^@]+}}@_ZN1SC1Ev
 // CHECK-SAME: (ptr noundef nonnull align 8 dereferenceable(40) [[THIS:%.*]]) unnamed_addr #[[ATTR2:[0-9]+]] comdat align 2 {
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[THIS_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8

 // CHECK-NEXT:    store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
 // CHECK-NEXT:    [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
 // CHECK-NEXT:    [[A:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], ptr [[THIS1]], i32 0, i32 0
 // CHECK-NEXT:    [[REF:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[THIS1]], i32 0, i32 2
 // CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[REF]], align 8
 // CHECK-NEXT:    [[PTR:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[THIS1]], i32 0, i32 1
 // CHECK-NEXT:    [[ARR:%.*]] = getelementptr inbounds [[STRUCT_S]], ptr [[THIS1]], i32 0, i32 3
 // CHECK-NEXT:    [[TMP1:%.*]] = getelementptr [4 x i32], ptr [[ARR]], i32 1
-// CHECK-NEXT:    [[TMP4:%.*]] = ptrtoint ptr [[TMP1]] to i64
-// CHECK-NEXT:    [[TMP5:%.*]] = ptrtoint ptr [[A]] to i64
-// CHECK-NEXT:    [[TMP6:%.*]] = sub i64 [[TMP4]], [[TMP5]]
-// CHECK-NEXT:    [[TMP7:%.*]] = sdiv exact i64 [[TMP6]], ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
+// CHECK-NEXT:    [[TMP2:%.*]] = ptrtoint ptr [[TMP1]] to i64
+// CHECK-NEXT:    [[TMP3:%.*]] = ptrtoint ptr [[A]] to i64
+// CHECK-NEXT:    [[TMP4:%.*]] = sub i64 [[TMP2]], [[TMP3]]
+// CHECK-NEXT:    [[TMP5:%.*]] = sdiv exact i64 [[TMP4]], ptrtoint (ptr getelementptr (i8, ptr null, i32 1) to i64)
 // CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes.1, i64 40, i1 false)
-// CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-NEXT:    store ptr [[THIS1]], ptr [[TMP9]], align 8
-// CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-NEXT:    [[TMP6:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-NEXT:    store ptr [[THIS1]], ptr [[TMP6]], align 8
+// CHECK-NEXT:    [[TMP7:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-NEXT:    store ptr [[A]], ptr [[TMP7]], align 8
+// CHECK-NEXT:    [[TMP8:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-NEXT:    store i64 [[TMP5]], ptr [[TMP8]], align 8
+// CHECK-NEXT:    [[TMP9:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
+// CHECK-NEXT:    store ptr null, ptr [[TMP9]], align 8
+// CHECK-NEXT:    [[TMP10:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
+// CHECK-NEXT:    store ptr [[THIS1]], ptr [[TMP10]], align 8
+// CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
 // CHECK-NEXT:    store ptr [[A]], ptr [[TMP11]], align 8
-// CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
-// CHECK-NEXT:    store i64 [[TMP7]], ptr [[TMP13]], align 8
-// CHECK-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
-// CHECK-NEXT:    store ptr null, ptr [[TMP14]], align 8
-// CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
-// CHECK-NEXT:    store ptr [[THIS1]], ptr [[TMP15]], align 8
-// CHECK-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
-// CHECK-NEXT:    store ptr [[A]], ptr [[TMP17]], align 8
-// CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
-// CHECK-NEXT:    store ptr null, ptr [[TMP19]], align 8
-// CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
-// CHECK-NEXT:    store ptr [[THIS1]], ptr [[TMP20]], align 8
-// CHECK-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
-// CHECK-NEXT:    store ptr [[TMP0]], ptr [[TMP22]], align 8
-// CHECK-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
-// CHECK-NEXT:    store ptr null, ptr [[TMP24]], align 8
-// CHECK-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 3
-// CHECK-NEXT:    store ptr [[THIS1]], ptr [[TMP25]], align 8
-// CHECK-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 3
-// CHECK-NEXT:    store ptr [[PTR]], ptr [[TMP27]], align 8
-// CHECK-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 3
-// CHECK-NEXT:    store ptr null, ptr [[TMP29]], align 8
-// CHECK-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 4
-// CHECK-NEXT:    store ptr [[THIS1]], ptr [[TMP30]], align 8
-// CHECK-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 4
-// CHECK-NEXT:    store ptr [[ARR]], ptr [[TMP32]], align 8
-// CHECK-NEXT:    [[TMP34:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 4
-// CHECK-NEXT:    store ptr null, ptr [[TMP34]], align 8
-// CHECK-NEXT:    [[TMP35:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
-// CHECK-NEXT:    [[TMP36:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
-// CHECK-NEXT:    [[TMP37:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
+// CHECK-NEXT:    [[TMP12:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
+// CHECK-NEXT:    store ptr null, ptr [[TMP12]], align 8
+// CHECK-NEXT:    [[TMP13:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
+// CHECK-NEXT:    store ptr [[THIS1]], ptr [[TMP13]], align 8
+// CHECK-NEXT:    [[TMP14:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
+// CHECK-NEXT:    store ptr [[TMP0]], ptr [[TMP14]], align 8
+// CHECK-NEXT:    [[TMP15:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
+// CHECK-NEXT:    store ptr null, ptr [[TMP15]], align 8
+// CHECK-NEXT:    [[TMP16:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 3
+// CHECK-NEXT:    store ptr [[THIS1]], ptr [[TMP16]], align 8
+// CHECK-NEXT:    [[TMP17:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 3
+// CHECK-NEXT:    store ptr [[PTR]], ptr [[TMP17]], align 8
+// CHECK-NEXT:    [[TMP18:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 3
+// CHECK-NEXT:    store ptr null, ptr [[TMP18]], align 8
+// CHECK-NEXT:    [[TMP19:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 4
+// CHECK-NEXT:    store ptr [[THIS1]], ptr [[TMP19]], align 8
+// CHECK-NEXT:    [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 4
+// CHECK-NEXT:    store ptr [[ARR]], ptr [[TMP20]], align 8
+// CHECK-NEXT:    [[TMP21:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 4
+// CHECK-NEXT:    store ptr null, ptr [[TMP21]], align 8
+// CHECK-NEXT:    [[TMP22:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
+// CHECK-NEXT:    [[TMP23:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
+// CHECK-NEXT:    [[TMP24:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
 // CHECK-NEXT:    [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
-// CHECK-NEXT:    [[TMP38:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
-// CHECK-NEXT:    store i32 1, ptr [[TMP38]], align 4
-// CHECK-NEXT:    [[TMP39:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
-// CHECK-NEXT:    store i32 5, ptr [[TMP39]], align 4
-// CHECK-NEXT:    [[TMP40:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
-// CHECK-NEXT:    store ptr [[TMP35]], ptr [[TMP40]], align 8
-// CHECK-NEXT:    [[TMP41:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
-// CHECK-NEXT:    store ptr [[TMP36]], ptr [[TMP41]], align 8
-// CHECK-NEXT:    [[TMP42:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
-// CHECK-NEXT:    store ptr [[TMP37]], ptr [[TMP42]], align 8
-// CHECK-NEXT:    [[TMP43:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
-// CHECK-NEXT:    store ptr @.offload_maptypes.2, ptr [[TMP43]], align 8
-// CHECK-NEXT:    [[TMP44:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
-// CHECK-NEXT:    store ptr null, ptr [[TMP44]], align 8
-// CHECK-NEXT:    [[TMP45:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
-// CHECK-NEXT:    store ptr null, ptr [[TMP45]], align 8
-// CHECK-NEXT:    [[TMP46:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
-// CHECK-NEXT:    store i64 0, ptr [[TMP46]], align 8
-// CHECK-NEXT:    [[TMP47:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN1S3fooEv_l14.region_id, ptr [[KERNEL_ARGS]])
-// CHECK-NEXT:    [[TMP48:%.*]] = icmp ne i32 [[TMP47]], 0
-// CHECK-NEXT:    br i1 [[TMP48]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
+// CHECK-NEXT:    [[TMP25:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
+// CHECK-NEXT:    store i32 1, ptr [[TMP25]], align 4
+// CHECK-NEXT:    [[TMP26:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
+// CHECK-NEXT:    store i32 5, ptr [[TMP26]], align 4
+// CHECK-NEXT:    [[TMP27:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
+// CHECK-NEXT:    store ptr [[TMP22]], ptr [[TMP27]], align 8
+// CHECK-NEXT:    [[TMP28:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
+// CHECK-NEXT:    store ptr [[TMP23]], ptr [[TMP28]], align 8
+// CHECK-NEXT:    [[TMP29:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
+// CHECK-NEXT:    store ptr [[TMP24]], ptr [[TMP29]], align 8
+// CHECK-NEXT:    [[TMP30:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
+// CHECK-NEXT:    store ptr @.offload_maptypes.2, ptr [[TMP30]], align 8
+// CHECK-NEXT:    [[TMP31:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
+// CHECK-NEXT:    store ptr null, ptr [[TMP31]], align 8
+// CHECK-NEXT:    [[TMP32:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
+// CHECK-NEXT:    store ptr null, ptr [[TMP32]], align 8
+// CHECK-NEXT:    [[TMP33:%.*]] = getelementptr inbounds [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
+// CHECK-NEXT:    store i64 0, ptr [[TMP33]], align 8
+// CHECK-NEXT:    [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 -1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN1S3fooEv_l14.region_id, ptr [[KERNEL_ARGS]])
+// CHECK-NEXT:    [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
+// CHECK-NEXT:    br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
 // CHECK:       omp_offload.failed:
 // CHECK-NEXT:    call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN1S3fooEv_l14(ptr [[THIS1]]) #[[ATTR5]]
 // CHECK-NEXT:    br label [[OMP_OFFLOAD_CONT]]
 // CHECK:       omp_offload.cont:
 // CHECK-NEXT:    ret void
 //
 //
 // CHECK-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}_main_l27

openmp/libomptarget/test/mapping/has_device_addr.cpp

#pragma omp target teams distribute parallel for has_device_addr(data_device[0])
     for (int i = 0; i < size; ++i)
       data_device[i] = i;
     omp_target_memcpy(data_host, data_device, bytes, 0, 0, host_id, device_id);
     for (int i = 0; i < size; ++i)
       assert(data_host[i] == i);
   }
 };
 
+void poo() {
+  short a = 1;
+  short &ar = a;
+
+  #pragma omp target data map(tofrom : ar) use_device_addr(ar)
+  {
+    #pragma omp target has_device_addr(ar)
+    {
+      ar = 222;
+      // CHECK: 222
+      printf("%hd %p\n", ar, &ar); // 222 p2
+    }
+  }
+  // CHECK: 222
+  printf("%hd %p\n", ar, &ar); // 222 p1
+}
+
+void noo() {
+  short *b = (short *)malloc(sizeof(short));
+  short *&br = b;
+  br = br - 1;
+
+ br[1] = 111;
+#pragma omp target data map(tofrom : br[1]) use_device_addr(br[1])
+#pragma omp target has_device_addr(br[1])
+  {
+    br[1] = 222;
+    // CHECK: 222
+    printf("%hd %p %p %p\n", br[1], br, &br[1], &br);
+  }
+  // CHECK: 222
+  printf("%hd %p %p %p\n", br[1], br, &br[1], &br);
+}
+
+void ooo() {
+  short a = 1;
+
+#pragma omp target data map(tofrom : a) use_device_addr(a)
+#pragma omp target has_device_addr(a)
+  {
+    a = 222;
+    // CHECK: 222
+    printf("%hd %p\n", a, &a);
+  }
+  // CHECK: 222
+  printf("%hd %p\n", a, &a);
+}
+
 int main() {
   view a;
   a.foo();
+  poo();
+  noo();
+  ooo();
   // CHECK: PASSED
   printf("PASSED\n");
 }