[OPENMP][NVPTX] Support memory coalescing for globalized variables.

Added support for memory coalescing for better performance for globalized variables. From now on all the globalized variables are represented as arrays of 32 elements and each thread accesses these elements using `tid & 31` as index. llvm-svn: 344049
2018-10-09 14:49:00 +00:00 · 2018-10-09 14:49:00 +00:00 · 9ea3c38597
parent f5fac1826a
commit 9ea3c38597
9 changed files with 175 additions and 71 deletions
--- a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp
@ -169,7 +169,7 @@ enum MachineConfiguration : unsigned {
  LaneIDMask = WarpSize - 1,

  /// Global memory alignment for performance.
-  GlobalMemoryAlignment = 256,
+  GlobalMemoryAlignment = 128,
 };

 enum NamedBarrier : unsigned {
@ -186,20 +186,30 @@ static bool stable_sort_comparator(const VarsDataTy P1, const VarsDataTy P2) {

 static RecordDecl *buildRecordForGlobalizedVars(
    ASTContext &C, ArrayRef<const ValueDecl *> EscapedDecls,
+    ArrayRef<const ValueDecl *> EscapedDeclsForTeams,
    llvm::SmallDenseMap<const ValueDecl *, const FieldDecl *>
        &MappedDeclsFields) {
-  if (EscapedDecls.empty())
+  if (EscapedDecls.empty() && EscapedDeclsForTeams.empty())
    return nullptr;
  SmallVector<VarsDataTy, 4> GlobalizedVars;
  for (const ValueDecl *D : EscapedDecls)
+    GlobalizedVars.emplace_back(
+        CharUnits::fromQuantity(std::max(
+            C.getDeclAlign(D).getQuantity(),
+            static_cast<CharUnits::QuantityType>(GlobalMemoryAlignment))),
+        D);
+  for (const ValueDecl *D : EscapedDeclsForTeams)
    GlobalizedVars.emplace_back(C.getDeclAlign(D), D);
  std::stable_sort(GlobalizedVars.begin(), GlobalizedVars.end(),
                   stable_sort_comparator);
  // Build struct _globalized_locals_ty {
-  //         /*  globalized vars  */
+  //         /*  globalized vars  */[32] align (max(decl_align, 128))
+  //         /*  globalized vars  */ for EscapedDeclsForTeams
  //       };
  RecordDecl *GlobalizedRD = C.buildImplicitRecord("_globalized_locals_ty");
  GlobalizedRD->startDefinition();
+  llvm::SmallPtrSet<const ValueDecl *, 16> SingleEscaped(
+      EscapedDeclsForTeams.begin(), EscapedDeclsForTeams.end());
  for (const auto &Pair : GlobalizedVars) {
    const ValueDecl *VD = Pair.second;
    QualType Type = VD->getType();
@ -208,19 +218,39 @@ static RecordDecl *buildRecordForGlobalizedVars(
    else
      Type = Type.getNonReferenceType();
    SourceLocation Loc = VD->getLocation();
-    auto *Field =
-        FieldDecl::Create(C, GlobalizedRD, Loc, Loc, VD->getIdentifier(), Type,
-                          C.getTrivialTypeSourceInfo(Type, SourceLocation()),
-                          /*BW=*/nullptr, /*Mutable=*/false,
-                          /*InitStyle=*/ICIS_NoInit);
-    Field->setAccess(AS_public);
-    GlobalizedRD->addDecl(Field);
-    if (VD->hasAttrs()) {
-      for (specific_attr_iterator<AlignedAttr> I(VD->getAttrs().begin()),
-           E(VD->getAttrs().end());
-           I != E; ++I)
-        Field->addAttr(*I);
+    FieldDecl *Field;
+    if (SingleEscaped.count(VD)) {
+      Field = FieldDecl::Create(
+          C, GlobalizedRD, Loc, Loc, VD->getIdentifier(), Type,
+          C.getTrivialTypeSourceInfo(Type, SourceLocation()),
+          /*BW=*/nullptr, /*Mutable=*/false,
+          /*InitStyle=*/ICIS_NoInit);
+      Field->setAccess(AS_public);
+      if (VD->hasAttrs()) {
+        for (specific_attr_iterator<AlignedAttr> I(VD->getAttrs().begin()),
+             E(VD->getAttrs().end());
+             I != E; ++I)
+          Field->addAttr(*I);
+      }
+    } else {
+      llvm::APInt ArraySize(32, WarpSize);
+      Type = C.getConstantArrayType(Type, ArraySize, ArrayType::Normal, 0);
+      Field = FieldDecl::Create(
+          C, GlobalizedRD, Loc, Loc, VD->getIdentifier(), Type,
+          C.getTrivialTypeSourceInfo(Type, SourceLocation()),
+          /*BW=*/nullptr, /*Mutable=*/false,
+          /*InitStyle=*/ICIS_NoInit);
+      Field->setAccess(AS_public);
+      llvm::APInt Align(32, std::max(C.getDeclAlign(VD).getQuantity(),
+                                     static_cast<CharUnits::QuantityType>(
+                                         GlobalMemoryAlignment)));
+      Field->addAttr(AlignedAttr::CreateImplicit(
+          C, AlignedAttr::GNU_aligned, /*IsAlignmentExpr=*/true,
+          IntegerLiteral::Create(C, Align,
+                                 C.getIntTypeForBitwidth(32, /*Signed=*/0),
+                                 SourceLocation())));
    }
+    GlobalizedRD->addDecl(Field);
    MappedDeclsFields.try_emplace(VD, Field);
  }
  GlobalizedRD->completeDefinition();
@ -344,7 +374,8 @@ class CheckVarsEscapingDeclContext final
    assert(!GlobalizedRD &&
           "Record for globalized variables is built already.");
    GlobalizedRD = ::buildRecordForGlobalizedVars(
-        CGF.getContext(), EscapedDecls.getArrayRef(), MappedDeclsFields);
+        CGF.getContext(), EscapedDecls.getArrayRef(), llvm::None,
+        MappedDeclsFields);
  }

 public:
@ -1849,8 +1880,7 @@ getDistributeLastprivateVars(const OMPExecutableDirective &D,
  }
  if (!Dir)
    return;
-  for (const OMPLastprivateClause *C :
-       Dir->getClausesOfKind<OMPLastprivateClause>()) {
+  for (const auto *C : Dir->getClausesOfKind<OMPLastprivateClause>()) {
    for (const Expr *E : C->getVarRefs()) {
      const auto *DE = cast<DeclRefExpr>(E->IgnoreParens());
      Vars.push_back(cast<ValueDecl>(DE->getDecl()->getCanonicalDecl()));
@ -1869,8 +1899,8 @@ llvm::Value *CGOpenMPRuntimeNVPTX::emitTeamsOutlinedFunction(
  if (getExecutionMode() == CGOpenMPRuntimeNVPTX::EM_SPMD) {
    getDistributeLastprivateVars(D, LastPrivates);
    if (!LastPrivates.empty())
-      GlobalizedRD = buildRecordForGlobalizedVars(
-          CGM.getContext(), LastPrivates, MappedDeclsFields);
+      GlobalizedRD = ::buildRecordForGlobalizedVars(
+          CGM.getContext(), llvm::None, LastPrivates, MappedDeclsFields);
  }

  // Emit target region as a standalone region.
@ -1899,9 +1929,9 @@ llvm::Value *CGOpenMPRuntimeNVPTX::emitTeamsOutlinedFunction(
        for (const auto &Pair : MappedDeclsFields) {
          assert(Pair.getFirst()->isCanonicalDecl() &&
                 "Expected canonical declaration");
-          Data.insert(std::make_pair(
-              Pair.getFirst(),
-              std::make_pair(Pair.getSecond(), Address::invalid())));
+          Data.insert(std::make_pair(Pair.getFirst(),
+                                     MappedVarData(Pair.getSecond(),
+                                                   /*IsOnePerTeam=*/true)));
        }
      }
      Rt.emitGenericVarsProlog(CGF, Loc);
@ -1935,18 +1965,20 @@ void CGOpenMPRuntimeNVPTX::emitGenericVarsProlog(CodeGenFunction &CGF,
  if (I == FunctionGlobalizedDecls.end())
    return;
  if (const RecordDecl *GlobalizedVarsRecord = I->getSecond().GlobalRecord) {
-    QualType RecTy = CGM.getContext().getRecordType(GlobalizedVarsRecord);
+    QualType GlobalRecTy = CGM.getContext().getRecordType(GlobalizedVarsRecord);

    // Recover pointer to this function's global record. The runtime will
    // handle the specifics of the allocation of the memory.
    // Use actual memory size of the record including the padding
    // for alignment purposes.
    unsigned Alignment =
-        CGM.getContext().getTypeAlignInChars(RecTy).getQuantity();
+        CGM.getContext().getTypeAlignInChars(GlobalRecTy).getQuantity();
    unsigned GlobalRecordSize =
-        CGM.getContext().getTypeSizeInChars(RecTy).getQuantity();
+        CGM.getContext().getTypeSizeInChars(GlobalRecTy).getQuantity();
    GlobalRecordSize = llvm::alignTo(GlobalRecordSize, Alignment);

+    llvm::PointerType *GlobalRecPtrTy =
+        CGF.ConvertTypeForMem(GlobalRecTy)->getPointerTo();
    llvm::Value *GlobalRecCastAddr;
    if (WithSPMDCheck ||
        getExecutionMode() == CGOpenMPRuntimeNVPTX::EM_Unknown) {
@ -1959,7 +1991,8 @@ void CGOpenMPRuntimeNVPTX::emitGenericVarsProlog(CodeGenFunction &CGF,
      // There is no need to emit line number for unconditional branch.
      (void)ApplyDebugLocation::CreateEmpty(CGF);
      CGF.EmitBlock(SPMDBB);
-      Address RecPtr = CGF.CreateMemTemp(RecTy, "_local_stack");
+      Address RecPtr = Address(llvm::ConstantPointerNull::get(GlobalRecPtrTy),
+                               CharUnits::fromQuantity(Alignment));
      CGF.EmitBranch(ExitBB);
      // There is no need to emit line number for unconditional branch.
      (void)ApplyDebugLocation::CreateEmpty(CGF);
@ -1974,9 +2007,9 @@ void CGOpenMPRuntimeNVPTX::emitGenericVarsProlog(CodeGenFunction &CGF,
                                  OMPRTL_NVPTX__kmpc_data_sharing_push_stack),
                              GlobalRecordSizeArg);
      GlobalRecCastAddr = Bld.CreatePointerBitCastOrAddrSpaceCast(
-          GlobalRecValue, CGF.ConvertTypeForMem(RecTy)->getPointerTo());
+          GlobalRecValue, GlobalRecPtrTy);
      CGF.EmitBlock(ExitBB);
-      auto *Phi = Bld.CreatePHI(GlobalRecCastAddr->getType(),
+      auto *Phi = Bld.CreatePHI(GlobalRecPtrTy,
                                /*NumReservedValues=*/2, "_select_stack");
      Phi->addIncoming(RecPtr.getPointer(), SPMDBB);
      Phi->addIncoming(GlobalRecCastAddr, NonSPMDBB);
@ -1994,12 +2027,12 @@ void CGOpenMPRuntimeNVPTX::emitGenericVarsProlog(CodeGenFunction &CGF,
                                  OMPRTL_NVPTX__kmpc_data_sharing_push_stack),
                              GlobalRecordSizeArg);
      GlobalRecCastAddr = Bld.CreatePointerBitCastOrAddrSpaceCast(
-          GlobalRecValue, CGF.ConvertTypeForMem(RecTy)->getPointerTo());
+          GlobalRecValue, GlobalRecPtrTy);
      I->getSecond().GlobalRecordAddr = GlobalRecValue;
      I->getSecond().IsInSPMDModeFlag = nullptr;
    }
    LValue Base =
-        CGF.MakeNaturalAlignPointeeAddrLValue(GlobalRecCastAddr, RecTy);
+        CGF.MakeNaturalAlignPointeeAddrLValue(GlobalRecCastAddr, GlobalRecTy);

    // Emit the "global alloca" which is a GEP from the global declaration
    // record using the pointer returned by the runtime.
@ -2012,9 +2045,34 @@ void CGOpenMPRuntimeNVPTX::emitGenericVarsProlog(CodeGenFunction &CGF,
            CGF.MakeAddrLValue(CGF.GetAddrOfLocalVar(VD), VD->getType());
        ParValue = CGF.EmitLoadOfScalar(ParLVal, Loc);
      }
-      const FieldDecl *FD = Rec.second.first;
-      LValue VarAddr = CGF.EmitLValueForField(Base, FD);
-      Rec.second.second = VarAddr.getAddress();
+      LValue VarAddr = CGF.EmitLValueForField(Base, Rec.second.FD);
+      // Emit VarAddr basing on lane-id if required.
+      QualType VarTy;
+      if (Rec.second.IsOnePerTeam) {
+        Rec.second.PrivateAddr = VarAddr.getAddress();
+        VarTy = Rec.second.FD->getType();
+      } else {
+        llvm::Value *Ptr = CGF.Builder.CreateInBoundsGEP(
+            VarAddr.getAddress().getPointer(),
+            {Bld.getInt32(0), getNVPTXLaneID(CGF)});
+        Rec.second.PrivateAddr =
+            Address(Ptr, CGM.getContext().getDeclAlign(Rec.first));
+        VarTy =
+            Rec.second.FD->getType()->castAsArrayTypeUnsafe()->getElementType();
+        VarAddr = CGF.MakeAddrLValue(Rec.second.PrivateAddr, VarTy,
+                                     AlignmentSource::Decl);
+      }
+      if (WithSPMDCheck ||
+                getExecutionMode() == CGOpenMPRuntimeNVPTX::EM_Unknown) {
+        assert(I->getSecond().IsInSPMDModeFlag &&
+               "Expected unknown execution mode or required SPMD check.");
+        Address GlobalPtr = Rec.second.PrivateAddr;
+        Address LocalAddr = CGF.CreateMemTemp(VarTy, Rec.second.FD->getName());
+        Rec.second.PrivateAddr = Address(
+            Bld.CreateSelect(I->getSecond().IsInSPMDModeFlag,
+                             LocalAddr.getPointer(), GlobalPtr.getPointer()),
+            LocalAddr.getAlignment());
+      }
      if (EscapedParam) {
        const auto *VD = cast<VarDecl>(Rec.first);
        CGF.EmitStoreOfScalar(ParValue, VarAddr);
@ -4047,7 +4105,7 @@ void CGOpenMPRuntimeNVPTX::emitFunctionProlog(CodeGenFunction &CGF,
  for (const ValueDecl *VD : VarChecker.getEscapedDecls()) {
    assert(VD->isCanonicalDecl() && "Expected canonical declaration");
    const FieldDecl *FD = VarChecker.getFieldForGlobalizedVar(VD);
-    Data.insert(std::make_pair(VD, std::make_pair(FD, Address::invalid())));
+    Data.insert(std::make_pair(VD, MappedVarData(FD)));
  }
  if (!NeedToDelayGlobalization) {
    emitGenericVarsProlog(CGF, D->getBeginLoc(), /*WithSPMDCheck=*/true);
@ -4074,7 +4132,7 @@ Address CGOpenMPRuntimeNVPTX::getAddressOfLocalVariable(CodeGenFunction &CGF,
    return Address::invalid();
  auto VDI = I->getSecond().LocalVarData.find(VD);
  if (VDI != I->getSecond().LocalVarData.end())
-    return VDI->second.second;
+    return VDI->second.PrivateAddr;
  if (VD->hasAttrs()) {
    for (specific_attr_iterator<OMPReferencedVarAttr> IT(VD->attr_begin()),
         E(VD->attr_end());
@ -4083,7 +4141,7 @@ Address CGOpenMPRuntimeNVPTX::getAddressOfLocalVariable(CodeGenFunction &CGF,
          cast<VarDecl>(cast<DeclRefExpr>(IT->getRef())->getDecl())
              ->getCanonicalDecl());
      if (VDI != I->getSecond().LocalVarData.end())
-        return VDI->second.second;
+        return VDI->second.PrivateAddr;
    }
  }
  return Address::invalid();
--- a/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.h
+++ b/clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.h
@ -373,9 +373,21 @@ private:
  llvm::Function *createParallelDataSharingWrapper(
      llvm::Function *OutlinedParallelFn, const OMPExecutableDirective &D);

+  /// The data for the single globalized variable.
+  struct MappedVarData {
+    /// Corresponding field in the global record.
+    const FieldDecl * FD = nullptr;
+    /// Corresponding address.
+    Address PrivateAddr = Address::invalid();
+    /// true, if only one element is required (for latprivates in SPMD mode),
+    /// false, if need to create based on the warp-size.
+    bool IsOnePerTeam = false;
+    MappedVarData() = delete;
+    MappedVarData(const FieldDecl *FD, bool IsOnePerTeam = false)
+        : FD(FD), IsOnePerTeam(IsOnePerTeam) {}
+  };
  /// The map of local variables to their addresses in the global memory.
-  using DeclToAddrMapTy = llvm::MapVector<const Decl *,
-      std::pair<const FieldDecl *, Address>>;
+  using DeclToAddrMapTy = llvm::MapVector<const Decl *, MappedVarData>;
  /// Set of the parameters passed by value escaping OpenMP context.
  using EscapedParamsTy = llvm::SmallPtrSet<const Decl *, 4>;
  struct FunctionData {
--- a/clang/test/OpenMP/declare_target_codegen_globalization.cpp
+++ b/clang/test/OpenMP/declare_target_codegen_globalization.cpp
@ -35,16 +35,21 @@ int maini1() {
 // CHECK-NOT: @__kmpc_data_sharing_push_stack

 // CHECK: define {{.*}}[[BAR]]()
-// CHECK: [[STACK:%.+]] = alloca [[GLOBAL_ST:%.+]],
+// CHECK: alloca i32,
+// CHECK: [[A_LOCAL_ADDR:%.+]] = alloca i32,
 // CHECK: [[RES:%.+]] = call i8 @__kmpc_is_spmd_exec_mode()
 // CHECK: [[IS_SPMD:%.+]] = icmp ne i8 [[RES]], 0
 // CHECK: br i1 [[IS_SPMD]], label
 // CHECK: br label
-// CHECK: [[RES:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i64 4, i16 0)
-// CHECK: [[GLOBALS:%.+]] = bitcast i8* [[RES]] to [[GLOBAL_ST]]*
+// CHECK: [[RES:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i64 128, i16 0)
+// CHECK: [[GLOBALS:%.+]] = bitcast i8* [[RES]] to [[GLOBAL_ST:%.+]]*
 // CHECK: br label
-// CHECK: [[ITEMS:%.+]] = phi [[GLOBAL_ST]]* [ [[STACK]], {{.+}} ], [ [[GLOBALS]], {{.+}} ]
+// CHECK: [[ITEMS:%.+]] = phi [[GLOBAL_ST]]* [ null, {{.+}} ], [ [[GLOBALS]], {{.+}} ]
 // CHECK: [[A_ADDR:%.+]] = getelementptr inbounds [[GLOBAL_ST]], [[GLOBAL_ST]]* [[ITEMS]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+// CHECK: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x()
+// CHECK: [[LID:%.+]] = and i32 [[TID]], 31
+// CHECK: [[A_GLOBAL_ADDR:%.+]] = getelementptr inbounds [32 x i32], [32 x i32]* [[A_ADDR]], i32 0, i32 [[LID]]
+// CHECK: [[A_ADDR:%.+]] = select i1 [[IS_SPMD]], i32* [[A_LOCAL_ADDR]], i32* [[A_GLOBAL_ADDR]]
 // CHECK: call {{.*}}[[FOO]](i32* dereferenceable{{.*}} [[A_ADDR]])
 // CHECK: br i1 [[IS_SPMD]], label
 // CHECK: [[BC:%.+]] = bitcast [[GLOBAL_ST]]* [[ITEMS]] to i8*
--- a/clang/test/OpenMP/nvptx_data_sharing.cpp
+++ b/clang/test/OpenMP/nvptx_data_sharing.cpp
@ -39,10 +39,16 @@ void test_ds(){
 // CK1: [[SHAREDARGS2:%.+]] = alloca i8**
 // CK1: call void @__kmpc_kernel_init
 // CK1: call void @__kmpc_data_sharing_init_stack
-// CK1: [[GLOBALSTACK:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i64 8, i16 0)
+// CK1: [[GLOBALSTACK:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i64 256, i16 0)
 // CK1: [[GLOBALSTACK2:%.+]] = bitcast i8* [[GLOBALSTACK]] to %struct._globalized_locals_ty*
-// CK1: [[A:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[GLOBALSTACK2]], i32 0, i32 0
-// CK1: [[B:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[GLOBALSTACK2]], i32 0, i32 1
+// CK1: [[A_ARR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[GLOBALSTACK2]], i32 0, i32 0
+// CK1: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x()
+// CK1: [[LID:%.+]] = and i32 [[TID]], 31
+// CK1: [[A:%.+]] = getelementptr inbounds [32 x i32], [32 x i32]* [[A_ARR]], i32 0, i32 [[LID]]
+// CK1: [[B_ARR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[GLOBALSTACK2]], i32 0, i32 1
+// CK1: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x()
+// CK1: [[LID:%.+]] = and i32 [[TID]], 31
+// CK1: [[B:%.+]] = getelementptr inbounds [32 x i32], [32 x i32]* [[B_ARR]], i32 0, i32 [[LID]]
 // CK1: store i32 10, i32* [[A]]
 // CK1: call void @__kmpc_kernel_prepare_parallel({{.*}}, i16 1)
 // CK1: call void @__kmpc_begin_sharing_variables(i8*** [[SHAREDARGS1]], i64 1)
--- a/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_distribute_parallel_generic_mode_codegen.cpp
@ -26,10 +26,13 @@ int main(int argc, char **argv) {
 // CHECK-LABEL: define internal void @__omp_offloading_{{.*}}_main_l17_worker(

 // CHECK: define weak void @__omp_offloading_{{.*}}_main_l17([10 x i32]* dereferenceable(40) %{{.+}}, [10 x i32]* dereferenceable(40) %{{.+}}, i32* dereferenceable(4) %{{.+}}, i{{64|32}} %{{.+}}, [10 x i32]* dereferenceable(40) %{{.+}})
-// CHECK: [[PTR:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i{{64|32}} 84, i16 0)
+// CHECK: [[PTR:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i{{64|32}} 2688, i16 0)
 // CHECK: [[STACK:%.+]] = bitcast i8* [[PTR]] to %struct._globalized_locals_ty*
 // CHECK: [[ARGC:%.+]] = load i32, i32* %{{.+}}, align
-// CHECK: [[ARGC_ADDR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[STACK]], i{{32|64}} 0, i{{32|64}} 0
+// CHECK: [[ARGC_ARR_ADDR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[STACK]], i{{32|64}} 0, i{{32|64}} 0
+// CHECK: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x()
+// CHECK: [[LID:%.+]] = and i32 [[TID]], 31
+// CHECK: [[ARGC_ADDR:%.+]] = getelementptr inbounds [32 x i32], [32 x i32]* [[ARGC_ARR_ADDR]], i32 0, i32 [[LID]]
 // CHECK: store i32 [[ARGC]], i32* [[ARGC_ADDR]],
 // CHECK: getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[STACK]], i{{32|64}} 0, i{{32|64}} 1
 // CHECK: getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[STACK]], i{{32|64}} 0, i{{32|64}} 2
--- a/clang/test/OpenMP/nvptx_parallel_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_parallel_codegen.cpp
@ -318,11 +318,14 @@ int bar(int n){
 // CHECK-32: [[A_ADDR:%.+]] = alloca i32,
 // CHECK-64: [[A_ADDR:%.+]] = alloca i64,
 // CHECK-64: [[CONV:%.+]] = bitcast i64* [[A_ADDR]] to i32*
-// CHECK: [[STACK:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i{{64|32}} 4, i16 0)
+// CHECK: [[STACK:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i{{64|32}} 128, i16 0)
 // CHECK: [[BC:%.+]] = bitcast i8* [[STACK]] to %struct._globalized_locals_ty*
 // CHECK-32: [[A:%.+]] = load i32, i32* [[A_ADDR]],
 // CHECK-64: [[A:%.+]] = load i32, i32* [[CONV]],
-// CHECK: [[GLOBAL_A_ADDR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[BC]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+// CHECK: [[GLOBAL_A_ADDR_ARR:%.+]] = getelementptr inbounds %struct._globalized_locals_ty, %struct._globalized_locals_ty* [[BC]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
+// CHECK: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x()
+// CHECK: [[LID:%.+]] = and i32 [[TID]], 31
+// CHECK: [[GLOBAL_A_ADDR:%.+]] = getelementptr inbounds [32 x i32], [32 x i32]* [[GLOBAL_A_ADDR_ARR]], i32 0, i32 [[LID]]
 // CHECK: store i32 [[A]], i32* [[GLOBAL_A_ADDR]],

 // CHECK-LABEL: define internal void @{{.+}}(i32* noalias %{{.+}}, i32* noalias %{{.+}}, i32* dereferenceable{{.*}})
--- a/clang/test/OpenMP/nvptx_target_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_target_codegen.cpp
@ -554,7 +554,8 @@ int baz(int f, double &a) {
  // CHECK: ret void

  // CHECK: define i32 [[BAZ]](i32 [[F:%.*]], double* dereferenceable{{.*}})
-  // CHECK: [[STACK:%.+]] = alloca [[GLOBAL_ST:%.+]],
+  // CHECK: alloca i32,
+  // CHECK: [[LOCAL_F_PTR:%.+]] = alloca i32,
  // CHECK: [[ZERO_ADDR:%.+]] = alloca i32,
  // CHECK: [[GTID:%.+]] = call i32 @__kmpc_global_thread_num(%struct.ident_t*
  // CHECK: store i32 0, i32* [[ZERO_ADDR]]
@ -562,11 +563,15 @@ int baz(int f, double &a) {
  // CHECK: [[IS_SPMD:%.+]] = icmp ne i8 [[RES]], 0
  // CHECK: br i1 [[IS_SPMD]], label
  // CHECK: br label
-  // CHECK: [[PTR:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i{{64|32}} 4, i16 0)
-  // CHECK: [[REC_ADDR:%.+]] = bitcast i8* [[PTR]] to [[GLOBAL_ST]]*
+  // CHECK: [[PTR:%.+]] = call i8* @__kmpc_data_sharing_push_stack(i{{64|32}} 128, i16 0)
+  // CHECK: [[REC_ADDR:%.+]] = bitcast i8* [[PTR]] to [[GLOBAL_ST:%.+]]*
  // CHECK: br label
-  // CHECK: [[ITEMS:%.+]] = phi [[GLOBAL_ST]]* [ [[STACK]], {{.+}} ], [ [[REC_ADDR]], {{.+}} ]
-  // CHECK: [[F_PTR:%.+]] = getelementptr inbounds [[GLOBAL_ST]], [[GLOBAL_ST]]* [[ITEMS]], i32 0, i32 0
+  // CHECK: [[ITEMS:%.+]] = phi [[GLOBAL_ST]]* [ null, {{.+}} ], [ [[REC_ADDR]], {{.+}} ]
+  // CHECK: [[F_PTR_ARR:%.+]] = getelementptr inbounds [[GLOBAL_ST]], [[GLOBAL_ST]]* [[ITEMS]], i32 0, i32 0
+  // CHECK: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x()
+  // CHECK: [[LID:%.+]] = and i32 [[TID]], 31
+  // CHECK: [[GLOBAL_F_PTR:%.+]] = getelementptr inbounds [32 x i32], [32 x i32]* [[F_PTR_ARR]], i32 0, i32 [[LID]]
+  // CHECK: [[F_PTR:%.+]] = select i1 [[IS_SPMD]], i32* [[LOCAL_F_PTR]], i32* [[GLOBAL_F_PTR]]
  // CHECK: store i32 %{{.+}}, i32* [[F_PTR]],

  // CHECK: [[RES:%.+]] = call i8 @__kmpc_is_spmd_exec_mode()
--- a/clang/test/OpenMP/nvptx_teams_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_teams_codegen.cpp
@ -36,10 +36,13 @@ int main (int argc, char **argv) {
 // CK1:  store {{.+}} 0, {{.+}},
 // CK1:  store i{{[0-9]+}} [[ARGC]], i{{[0-9]+}}* [[ARGCADDR]],
 // CK1-64:  [[CONV:%.+]] = bitcast i{{[0-9]+}}* [[ARGCADDR]] to i{{[0-9]+}}*
-// CK1:  call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} 4, i16 0)
+// CK1:  call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} 128, i16 0)
 // CK1-64:  [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[CONV]]
 // CK1-32:  [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[ARGCADDR]]
-// CK1:  [[ARGCADDR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
+// CK1:  [[ARGCADDR_ARR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
+// CK1:  [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x()
+// CK1:  [[LID:%.+]] = and i32 [[TID]], 31
+// CK1:  [[ARGCADDR:%.+]] = getelementptr inbounds [32 x i32], [32 x i32]* [[ARGCADDR_ARR]], i32 0, i32 [[LID]]
 // CK1:  store i{{[0-9]+}} [[ARG]], i{{[0-9]+}}* [[ARGCADDR]],
 // CK1:  store i{{[0-9]+}}* [[ARGCADDR]], i{{[0-9]+}}** [[ARGCADDR_PTR]],
 // CK1:  [[ARGCADDR_PTR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[ARGCADDR_PTR]],
@ -53,9 +56,12 @@ int main (int argc, char **argv) {
 // CK1: [[ARGCADDR_PTR:%.+]] = alloca i{{.+}}***,
 // CK1: [[ARGCADDR:%.+]] = alloca i{{.+}}**,
 // CK1: store i{{.+}}** [[ARGC]], i{{.+}}*** [[ARGCADDR]]
-// CK1: call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} {{4|8}}, i16 0)
+// CK1: call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} {{128|256}}, i16 0)
 // CK1: [[ARG:%.+]] = load i{{[0-9]+}}**, i{{[0-9]+}}*** [[ARGCADDR]]
-// CK1: [[ARGCADDR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
+// CK1: [[ARGCADDR_ARR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
+// CK1: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x()
+// CK1: [[LID:%.+]] = and i32 [[TID]], 31
+// CK1: [[ARGCADDR:%.+]] = getelementptr inbounds [32 x i8**], [32 x i8**]* [[ARGCADDR_ARR]], i32 0, i32 [[LID]]
 // CK1: store i{{[0-9]+}}** [[ARG]], i{{[0-9]+}}*** [[ARGCADDR]],
 // CK1: store i8*** [[ARGCADDR]], i8**** [[ARGCADDR_PTR]],
 // CK1: [[ARGCADDR_PTR_REF:%.+]] = load i{{.+}}**, i{{.+}}*** [[ARGCADDR_PTR]],
@ -111,10 +117,13 @@ int main (int argc, char **argv) {
 // CK2-64: [[ACONV:%.+]] = bitcast i64* [[AADDR]] to i32*
 // CK2-64: [[BCONV:%.+]] = bitcast i64* [[BADDR]] to i32*
 // CK2-64: [[CONV:%.+]] = bitcast i64* [[ARGCADDR]] to i32*
-// CK2:  call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} 4, i16 0)
+// CK2:  call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} 128, i16 0)
 // CK2-64:  [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[CONV]]
 // CK2-32:  [[ARG:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[ARGCADDR]]
-// CK2:  [[ARGCADDR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
+// CK2:  [[ARGCADDR_ARR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
+// CK2:  [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x()
+// CK2:  [[LID:%.+]] = and i32 [[TID]], 31
+// CK2:  [[ARGCADDR:%.+]] = getelementptr inbounds [32 x i32], [32 x i32]* [[ARGCADDR_ARR]], i32 0, i32 [[LID]]
 // CK2:  store i{{[0-9]+}} [[ARG]], i{{[0-9]+}}* [[ARGCADDR]],
 // CK2:  {{%.+}} = call i32 @__kmpc_global_thread_num(
 // CK2:  store i{{[0-9]+}}* [[ARGCADDR]], i{{[0-9]+}}** [[ARGCADDR_PTR]],
@ -132,9 +141,12 @@ int main (int argc, char **argv) {
 // CK2: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[AADDR]],
 // CK2: store i{{[0-9]+}} [[B_IN]], i{{[0-9]+}}* [[BADDR]],
 // CK2: store i{{[0-9]+}}** [[ARGC]], i{{[0-9]+}}*** [[ARGCADDR]],
-// CK2: call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} {{4|8}}, i16 0)
+// CK2: call i8* @__kmpc_data_sharing_push_stack(i{{[0-9]+}} {{128|256}}, i16 0)
 // CK2: [[ARG:%.+]] = load i{{[0-9]+}}**, i{{[0-9]+}}*** [[ARGCADDR]]
-// CK2: [[ARGCADDR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
+// CK2: [[ARGCADDR_ARR:%.+]] = getelementptr inbounds %struct.{{.*}}, %struct.{{.*}}* %{{.*}}, i{{[0-9]+}} 0, i{{[0-9]+}} 0
+// CK2: [[TID:%.+]] = call i32 @llvm.nvvm.read.ptx.sreg.tid.x()
+// CK2: [[LID:%.+]] = and i32 [[TID]], 31
+// CK2: [[ARGCADDR:%.+]] = getelementptr inbounds [32 x i8**], [32 x i8**]* [[ARGCADDR_ARR]], i32 0, i32 [[LID]]
 // CK2: store i{{[0-9]+}}** [[ARG]], i{{[0-9]+}}*** [[ARGCADDR]],
 // CK2: {{%.+}} = call i32 @__kmpc_global_thread_num(
 // CK2: store i{{[0-9]+}}*** [[ARGCADDR]], i{{[0-9]+}}**** [[ARGCADDR_PTR]],
--- a/clang/test/OpenMP/nvptx_teams_reduction_codegen.cpp
+++ b/clang/test/OpenMP/nvptx_teams_reduction_codegen.cpp
@ -597,9 +597,9 @@ int bar(int n){
  // CHECK: [[OF:%.+]] = mul nuw i[[SZ]] [[NUM_TEAMS]], 1
  // CHECK: [[POS1:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[OF]]
  // CHECK: [[POS2:%.+]] = sub nuw i[[SZ]] [[POS1]], 1
-  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 256
+  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 128
  // CHECK: [[POS4:%.+]] = add nuw i[[SZ]] [[POS3]], 1
-  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 256
+  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 128
  //
  // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 1
  // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]],
@ -643,9 +643,9 @@ int bar(int n){
  // CHECK: [[OF:%.+]] = mul nuw i[[SZ]] [[NUM_TEAMS]], 1
  // CHECK: [[POS1:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[OF]]
  // CHECK: [[POS2:%.+]] = sub nuw i[[SZ]] [[POS1]], 1
-  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 256
+  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 128
  // CHECK: [[POS4:%.+]] = add nuw i[[SZ]] [[POS3]], 1
-  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 256
+  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 128
  //
  // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 4, [[TEAM]]
  // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD_NEXT]], [[P]]
@ -1024,9 +1024,9 @@ int bar(int n){
  // CHECK: [[OF:%.+]] = mul nuw i[[SZ]] [[NUM_TEAMS]], 4
  // CHECK: [[POS1:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[OF]]
  // CHECK: [[POS2:%.+]] = sub nuw i[[SZ]] [[POS1]], 1
-  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 256
+  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 128
  // CHECK: [[POS4:%.+]] = add nuw i[[SZ]] [[POS3]], 1
-  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 256
+  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 128
  //
  // CHECK: [[ELT_REF:%.+]] = getelementptr inbounds [[RLT]], [[RLT]]* [[RED_LIST]], i[[SZ]] 0, i[[SZ]] 1
  // CHECK: [[ELT_VOID:%.+]] = load i8*, i8** [[ELT_REF]],
@ -1072,9 +1072,9 @@ int bar(int n){
  // CHECK: [[OF:%.+]] = mul nuw i[[SZ]] [[NUM_TEAMS]], 4
  // CHECK: [[POS1:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD]], [[OF]]
  // CHECK: [[POS2:%.+]] = sub nuw i[[SZ]] [[POS1]], 1
-  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 256
+  // CHECK: [[POS3:%.+]] = udiv i[[SZ]] [[POS2]], 128
  // CHECK: [[POS4:%.+]] = add nuw i[[SZ]] [[POS3]], 1
-  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 256
+  // CHECK: [[SCRATCHPAD_NEXT:%.+]] = mul nuw i[[SZ]] [[POS4]], 128
  //
  // CHECK: [[P:%.+]] = mul nuw i[[SZ]] 2, [[TEAM]]
  // CHECK: [[SCRATCHPAD_ELT_PTR64:%.+]] = add nuw i[[SZ]] [[SCRATCHPAD_NEXT]], [[P]]