[OpenMP] Add NoSync attributes to alloc / free shared RTL calls

[jhuber6]

Summary

This patch adds the nosync attribute to the __kmpc_alloc_shared and
__kmpc_free_shared runtime library calls. This allows code analysis to
know that these functins dont contain any barriers. This will help
optimizations reason about the CFG of blocks containing these calls.

[jdoerfert]

This patch adds the nosync attribute to the kmpc_alloc_shared and
kmpc_free_shared runtime library calls. This allows code analysis to
know that these functins dont contain any barriers. This will help
optimizations reason about the CFG of blocks containing these calls.

Technically they do contain synchronization events but neither the user nor we use them to synchronize
things not inside the functions. So, from the outside perspective they are nosync.

LG

[xbolva00]

llvm/include/llvm/Frontend/OpenMP/OMPKinds.def

545

AttributeSet()?

[jhuber6]

llvm/include/llvm/Frontend/OpenMP/OMPKinds.def

545

We want this without optimistic attributes set. Optimistic attributes will be the default soon anyway.

llvm/include/llvm/Frontend/OpenMP/OMPKinds.def

__OMP_ATTRS_SET(ForkAttrs, OptimisticAttributes
                                : AttributeSet(EnumAttr(NoUnwind)))
 
 __OMP_ATTRS_SET(ReadOnlyPtrAttrs,
                 OptimisticAttributes
                     ? AttributeSet(EnumAttr(ReadOnly), EnumAttr(NoFree),
                                    EnumAttr(NoCapture))
                     : AttributeSet())
 
+__OMP_ATTRS_SET(DeviceAllocAttrs,
+                OptimisticAttributes
+                    ? AttributeSet(EnumAttr(NoUnwind), EnumAttr(NoSync))
+                    : AttributeSet(EnumAttr(NoUnwind), EnumAttr(NoSync)))
+
 #if 0
 __OMP_ATTRS_SET(WriteOnlyPtrAttrs,
                 OptimisticAttributes
                     ? AttributeSet(EnumAttr(WriteOnly), EnumAttr(NoFree),
                                    EnumAttr(NoCapture))
                     : AttributeSet())
 #endif
 
__OMP_RTL_ATTRS(__kmpc_doacross_init, BarrierAttrs, AttributeSet(),
                 ParamAttrs(ReadOnlyPtrAttrs))
 __OMP_RTL_ATTRS(__kmpc_doacross_post, BarrierAttrs, AttributeSet(),
                 ParamAttrs(ReadOnlyPtrAttrs, AttributeSet(), ReadOnlyPtrAttrs))
 __OMP_RTL_ATTRS(__kmpc_doacross_wait, BarrierAttrs, AttributeSet(),
                 ParamAttrs(ReadOnlyPtrAttrs, AttributeSet(), ReadOnlyPtrAttrs))
 __OMP_RTL_ATTRS(__kmpc_doacross_fini, BarrierAttrs, AttributeSet(),
                 ParamAttrs(ReadOnlyPtrAttrs))
 
-__OMP_RTL_ATTRS(__kmpc_alloc_shared, DefaultAttrs, ReturnPtrAttrs,
+__OMP_RTL_ATTRS(__kmpc_alloc_shared, DeviceAllocAttrs, ReturnPtrAttrs,
                 ParamAttrs())
-__OMP_RTL_ATTRS(__kmpc_free_shared, AllocAttrs, AttributeSet(),
+__OMP_RTL_ATTRS(__kmpc_free_shared, DeviceAllocAttrs, AttributeSet(),
                 ParamAttrs(NoCaptureAttrs))
 
 __OMP_RTL_ATTRS(__kmpc_alloc, DefaultAttrs, ReturnPtrAttrs, ParamAttrs())
 __OMP_RTL_ATTRS(__kmpc_free, AllocAttrs, AttributeSet(), ParamAttrs())
 
 __OMP_RTL_ATTRS(__kmpc_init_allocator, DefaultAttrs, ReturnPtrAttrs,
                 ParamAttrs())
 __OMP_RTL_ATTRS(__kmpc_destroy_allocator, AllocAttrs, AttributeSet(),

llvm/test/Transforms/OpenMP/remove_globalization.ll

-; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-globals
 ; RUN: opt -S -passes=openmp-opt < %s | FileCheck %s
 ; RUN: opt -passes=openmp-opt -pass-remarks=openmp-opt -pass-remarks-missed=openmp-opt -disable-output < %s 2>&1 | FileCheck %s -check-prefix=CHECK-REMARKS
 ; RUN: opt -openmp-opt-disable-deglobalization -S -passes=openmp-opt < %s | FileCheck %s --check-prefix=CHECK-DISABLED
 target datalayout = "e-i64:64-i128:128-v16:16-v32:32-n16:32:64"
 target triple = "nvptx64"
 
 ; CHECK-REMARKS: remark: remove_globalization.c:4:2: Could not move globalized variable to the stack. Variable is potentially captured in call. Mark parameter as `__attribute__((noescape))` to override.
 ; CHECK-REMARKS: remark: remove_globalization.c:2:2: Moving globalized variable to the stack.
 ; CHECK-REMARKS: remark: remove_globalization.c:6:2: Moving globalized variable to the stack.
 ; CHECK-REMARKS: remark: remove_globalization.c:4:2: Found thread data sharing on the GPU. Expect degraded performance due to data globalization.
 
 @S = external local_unnamed_addr global i8*
 
 %struct.ident_t = type { i32, i32, i32, i32, i8* }
 
 declare i32 @__kmpc_target_init(%struct.ident_t*, i1, i1, i1)
 declare void @__kmpc_target_deinit(%struct.ident_t*, i1, i1)
 
+;.
+; CHECK: @[[S:[a-zA-Z0-9_$"\\.-]+]] = external local_unnamed_addr global i8*
+;.
+; CHECK-DISABLED: @[[S:[a-zA-Z0-9_$"\\.-]+]] = external local_unnamed_addr global i8*
+;.
 define void @kernel() {
 ; CHECK-LABEL: define {{[^@]+}}@kernel() {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* nonnull null, i1 false, i1 false, i1 true)
-; CHECK-NEXT:    call void @foo() #[[ATTR0:[0-9]+]]
-; CHECK-NEXT:    call void @bar() #[[ATTR0]]
+; CHECK-NEXT:    call void @foo() #[[ATTR4:[0-9]+]]
+; CHECK-NEXT:    call void @bar() #[[ATTR4]]
 ; CHECK-NEXT:    call void @unknown_no_openmp()
 ; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* nonnull null, i1 false, i1 true)
 ; CHECK-NEXT:    ret void
 ;
 ; CHECK-DISABLED-LABEL: define {{[^@]+}}@kernel() {
 ; CHECK-DISABLED-NEXT:  entry:
 ; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* nonnull null, i1 false, i1 false, i1 true)
-; CHECK-DISABLED-NEXT:    call void @foo() #[[ATTR0:[0-9]+]]
-; CHECK-DISABLED-NEXT:    call void @bar() #[[ATTR0]]
+; CHECK-DISABLED-NEXT:    call void @foo() #[[ATTR4:[0-9]+]]
+; CHECK-DISABLED-NEXT:    call void @bar() #[[ATTR4]]
 ; CHECK-DISABLED-NEXT:    call void @unknown_no_openmp()
 ; CHECK-DISABLED-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* nonnull null, i1 false, i1 true)
 ; CHECK-DISABLED-NEXT:    ret void
 ;
 entry:
   %0 = call i32 @__kmpc_target_init(%struct.ident_t* nonnull null, i1 false, i1 true, i1 true)
   call void @foo()
   call void @bar()
   call void @unknown_no_openmp()
   call void @__kmpc_target_deinit(%struct.ident_t* nonnull null, i1 false, i1 true)
   ret void
 }
 
 define internal void @foo() {
 ; CHECK-LABEL: define {{[^@]+}}@foo
-; CHECK-SAME: () #[[ATTR0]] {
+; CHECK-SAME: () #[[ATTR0:[0-9]+]] {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[TMP0:%.*]] = alloca i8, i64 4, align 1
 ; CHECK-NEXT:    ret void
 ;
 ; CHECK-DISABLED-LABEL: define {{[^@]+}}@foo
-; CHECK-DISABLED-SAME: () #[[ATTR0]] {
+; CHECK-DISABLED-SAME: () #[[ATTR0:[0-9]+]] {
 ; CHECK-DISABLED-NEXT:  entry:
 ; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = alloca i8, i64 4, align 1
 ; CHECK-DISABLED-NEXT:    ret void
 ;
 entry:
   %0 = call i8* @__kmpc_alloc_shared(i64 4), !dbg !12
   call void @use(i8* %0)
   call void @__kmpc_free_shared(i8* %0, i64 4)
   ret void
 }
 
 define internal void @bar() {
 ; CHECK-LABEL: define {{[^@]+}}@bar
 ; CHECK-SAME: () #[[ATTR0]] {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[TMP0:%.*]] = call i8* @__kmpc_alloc_shared(i64 noundef 4) #[[ATTR0]], !dbg [[DBG8:![0-9]+]]
-; CHECK-NEXT:    call void @share(i8* nofree writeonly [[TMP0]]) #[[ATTR4:[0-9]+]]
-; CHECK-NEXT:    call void @__kmpc_free_shared(i8* [[TMP0]], i64 noundef 4) #[[ATTR0]]
+; CHECK-NEXT:    [[TMP0:%.*]] = call i8* @__kmpc_alloc_shared(i64 noundef 4) #[[ATTR4]], !dbg [[DBG8:![0-9]+]]
+; CHECK-NEXT:    call void @share(i8* nofree writeonly [[TMP0]]) #[[ATTR5:[0-9]+]]
+; CHECK-NEXT:    call void @__kmpc_free_shared(i8* [[TMP0]], i64 noundef 4) #[[ATTR4]]
 ; CHECK-NEXT:    ret void
 ;
 ; CHECK-DISABLED-LABEL: define {{[^@]+}}@bar
 ; CHECK-DISABLED-SAME: () #[[ATTR0]] {
 ; CHECK-DISABLED-NEXT:  entry:
-; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i8* @__kmpc_alloc_shared(i64 noundef 4) #[[ATTR0]], !dbg [[DBG8:![0-9]+]]
-; CHECK-DISABLED-NEXT:    call void @share(i8* nofree writeonly [[TMP0]]) #[[ATTR4:[0-9]+]]
-; CHECK-DISABLED-NEXT:    call void @__kmpc_free_shared(i8* [[TMP0]], i64 noundef 4) #[[ATTR0]]
+; CHECK-DISABLED-NEXT:    [[TMP0:%.*]] = call i8* @__kmpc_alloc_shared(i64 noundef 4) #[[ATTR4]], !dbg [[DBG8:![0-9]+]]
+; CHECK-DISABLED-NEXT:    call void @share(i8* nofree writeonly [[TMP0]]) #[[ATTR5:[0-9]+]]
+; CHECK-DISABLED-NEXT:    call void @__kmpc_free_shared(i8* [[TMP0]], i64 noundef 4) #[[ATTR4]]
 ; CHECK-DISABLED-NEXT:    ret void
 ;
 entry:
   %0 = call i8* @__kmpc_alloc_shared(i64 4), !dbg !13
   call void @share(i8* %0)
   call void @__kmpc_free_shared(i8* %0, i64 4)
   ret void
 }

 !7 = !{i32 7, !"openmp-device", i32 50}
 !8 = distinct !DISubprogram(name: "foo", scope: !1, file: !1, line: 1, type: !11, scopeLine: 1, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
 !9 = distinct !DISubprogram(name: "bar", scope: !1, file: !1, line: 1, type: !11, scopeLine: 1, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
 !10 = distinct !DISubprogram(name: "unused", scope: !1, file: !1, line: 1, type: !11, scopeLine: 1, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
 !11 = !DISubroutineType(types: !2)
 !12 = !DILocation(line: 2, column: 2, scope: !8)
 !13 = !DILocation(line: 4, column: 2, scope: !9)
 !14 = !DILocation(line: 6, column: 2, scope: !9)
+;.
+; CHECK: attributes #[[ATTR0]] = { nosync nounwind }
+; CHECK: attributes #[[ATTR1]] = { nofree nosync nounwind readnone willreturn }
+; CHECK: attributes #[[ATTR2]] = { nofree nosync nounwind willreturn writeonly }
+; CHECK: attributes #[[ATTR3:[0-9]+]] = { "llvm.assume"="omp_no_openmp" }
+; CHECK: attributes #[[ATTR4]] = { nounwind }
+; CHECK: attributes #[[ATTR5]] = { nosync nounwind writeonly }
+;.
+; CHECK-DISABLED: attributes #[[ATTR0]] = { nosync nounwind }
+; CHECK-DISABLED: attributes #[[ATTR1]] = { nofree nosync nounwind readnone willreturn }
+; CHECK-DISABLED: attributes #[[ATTR2]] = { nofree nosync nounwind willreturn writeonly }
+; CHECK-DISABLED: attributes #[[ATTR3:[0-9]+]] = { "llvm.assume"="omp_no_openmp" }
+; CHECK-DISABLED: attributes #[[ATTR4]] = { nounwind }
+; CHECK-DISABLED: attributes #[[ATTR5]] = { nosync nounwind writeonly }
+;.
+; CHECK: [[META0:![0-9]+]] = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 13.0.0", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, splitDebugInlining: false, nameTableKind: None)
+; CHECK: [[META1:![0-9]+]] = !DIFile(filename: "remove_globalization.c", directory: "/tmp/remove_globalization.c")
+; CHECK: [[META2:![0-9]+]] = !{}
+; CHECK: [[META3:![0-9]+]] = !{i32 2, !"Debug Info Version", i32 3}
+; CHECK: [[META4:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+; CHECK: [[META5:![0-9]+]] = !{i32 7, !"openmp", i32 50}
+; CHECK: [[META6:![0-9]+]] = !{i32 7, !"openmp-device", i32 50}
+; CHECK: [[META7:![0-9]+]] = !{void ()* @kernel, !"kernel", i32 1}
+; CHECK: [[DBG8]] = !DILocation(line: 4, column: 2, scope: !9)
+; CHECK: [[META9:![0-9]+]] = distinct !DISubprogram(name: "bar", scope: !1, file: !1, line: 1, type: !10, scopeLine: 1, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
+; CHECK: [[META10:![0-9]+]] = !DISubroutineType(types: !2)
+;.
+; CHECK-DISABLED: [[META0:![0-9]+]] = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 13.0.0", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, splitDebugInlining: false, nameTableKind: None)
+; CHECK-DISABLED: [[META1:![0-9]+]] = !DIFile(filename: "remove_globalization.c", directory: "/tmp/remove_globalization.c")
+; CHECK-DISABLED: [[META2:![0-9]+]] = !{}
+; CHECK-DISABLED: [[META3:![0-9]+]] = !{i32 2, !"Debug Info Version", i32 3}
+; CHECK-DISABLED: [[META4:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
+; CHECK-DISABLED: [[META5:![0-9]+]] = !{i32 7, !"openmp", i32 50}
+; CHECK-DISABLED: [[META6:![0-9]+]] = !{i32 7, !"openmp-device", i32 50}
+; CHECK-DISABLED: [[META7:![0-9]+]] = !{void ()* @kernel, !"kernel", i32 1}
+; CHECK-DISABLED: [[DBG8]] = !DILocation(line: 4, column: 2, scope: !9)
+; CHECK-DISABLED: [[META9:![0-9]+]] = distinct !DISubprogram(name: "bar", scope: !1, file: !1, line: 1, type: !10, scopeLine: 1, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
+; CHECK-DISABLED: [[META10:![0-9]+]] = !DISubroutineType(types: !2)
+; CHECK-DISABLED: [[DBG11]] = !DILocation(line: 6, column: 2, scope: !9)
+;.

llvm/test/Transforms/OpenMP/replace_globalization.ll

-; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-globals --include-generated-funcs
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --function-signature --check-attributes --check-globals --include-generated-funcs
 ; RUN: opt -S -passes='openmp-opt' < %s | FileCheck %s
 ; RUN: opt -passes=openmp-opt -pass-remarks=openmp-opt -disable-output < %s 2>&1 | FileCheck %s -check-prefix=CHECK-REMARKS
 target datalayout = "e-i64:64-i128:128-v16:16-v32:32-n16:32:64"
 target triple = "nvptx64"
 
+; CHECK-REMARKS: remark: replace_globalization.c:5:7: Replaced globalized variable with 16 bytes of shared memory
+; CHECK-REMARKS: remark: replace_globalization.c:5:14: Replaced globalized variable with 4 bytes of shared memory
+; CHECK-REMARKS-NOT: 6 bytes
+
 %struct.ident_t = type { i32, i32, i32, i32, i8* }
 
 @S = external local_unnamed_addr global i8*
 @0 = private unnamed_addr constant [113 x i8] c";llvm/test/Transforms/OpenMP/custom_state_machines_remarks.c;__omp_offloading_2a_d80d3d_test_fallback_l11;11;1;;\00", align 1
 @1 = private unnamed_addr constant %struct.ident_t { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([113 x i8], [113 x i8]* @0, i32 0, i32 0) }, align 8
 
-; CHECK-REMARKS: remark: replace_globalization.c:5:7: Replaced globalized variable with 16 bytes of shared memory
-; CHECK-REMARKS: remark: replace_globalization.c:5:14: Replaced globalized variable with 4 bytes of shared memory
-; CHECK-REMARKS-NOT: 6 bytes
 
 define dso_local void @foo() {
 entry:
   %c = call i32 @__kmpc_target_init(%struct.ident_t* @1, i1 false, i1 true, i1 true)
   %x = call i8* @__kmpc_alloc_shared(i64 4)
   call void @unknown_no_openmp()
   %x_on_stack = bitcast i8* %x to i32*
   %0 = bitcast i32* %x_on_stack to i8*

 ; CHECK: @[[GLOB0:[0-9]+]] = private unnamed_addr constant [113 x i8] c"
 ; CHECK: @[[GLOB1:[0-9]+]] = private unnamed_addr constant [[STRUCT_IDENT_T:%.*]] { i32 0, i32 2, i32 0, i32 0, i8* getelementptr inbounds ([113 x i8], [113 x i8]* @[[GLOB0]], i32 0, i32 0) }, align 8
 ; CHECK: @[[X:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [16 x i8] undef, align 32
 ; CHECK: @[[Y:[a-zA-Z0-9_$"\\.-]+]] = internal addrspace(3) global [4 x i8] undef, align 32
 ;.
 ; CHECK-LABEL: define {{[^@]+}}@foo() {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[C:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 false, i1 true)
-; CHECK-NEXT:    [[X:%.*]] = call i8* @__kmpc_alloc_shared(i64 4) #[[ATTR1:[0-9]+]]
+; CHECK-NEXT:    [[X:%.*]] = call i8* @__kmpc_alloc_shared(i64 4) #[[ATTR4:[0-9]+]]
 ; CHECK-NEXT:    call void @unknown_no_openmp()
-; CHECK-NEXT:    call void @use.internalized(i8* nofree writeonly [[X]]) #[[ATTR4:[0-9]+]]
-; CHECK-NEXT:    call void @__kmpc_free_shared(i8* [[X]], i64 4) #[[ATTR1]]
+; CHECK-NEXT:    call void @use.internalized(i8* nofree writeonly [[X]]) #[[ATTR5:[0-9]+]]
+; CHECK-NEXT:    call void @__kmpc_free_shared(i8* [[X]], i64 4) #[[ATTR4]]
 ; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
 ; CHECK-NEXT:    ret void
 ;
 ;
 ; CHECK-LABEL: define {{[^@]+}}@bar() {
 ; CHECK-NEXT:    [[C:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* @[[GLOB1]], i1 false, i1 true, i1 true)
 ; CHECK-NEXT:    call void @unknown_no_openmp()
 ; CHECK-NEXT:    call void @baz()
 ; CHECK-NEXT:    call void @qux()
 ; CHECK-NEXT:    call void @negative_qux_spmd()
 ; CHECK-NEXT:    call void @__kmpc_target_deinit(%struct.ident_t* @[[GLOB1]], i1 false, i1 true)
 ; CHECK-NEXT:    ret void
 ;
 ;
 ; CHECK-LABEL: define {{[^@]+}}@baz() {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[CALL:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* noundef nonnull @[[GLOB1]], i1 noundef false, i1 noundef false, i1 noundef true)
 ; CHECK-NEXT:    call void @unknown_no_openmp()
 ; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[CALL]], -1
 ; CHECK-NEXT:    br i1 [[CMP]], label [[MASTER:%.*]], label [[EXIT:%.*]]
 ; CHECK:       master:
-; CHECK-NEXT:    call void @use.internalized(i8* nofree writeonly addrspacecast (i8 addrspace(3)* getelementptr inbounds ([16 x i8], [16 x i8] addrspace(3)* @x, i32 0, i32 0) to i8*)) #[[ATTR4]]
+; CHECK-NEXT:    call void @use.internalized(i8* nofree writeonly addrspacecast (i8 addrspace(3)* getelementptr inbounds ([16 x i8], [16 x i8] addrspace(3)* @x, i32 0, i32 0) to i8*)) #[[ATTR5]]
 ; CHECK-NEXT:    br label [[EXIT]]
 ; CHECK:       exit:
 ; CHECK-NEXT:    ret void
 ;
 ;
 ; CHECK-LABEL: define {{[^@]+}}@qux() {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[CALL:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* noundef nonnull @[[GLOB1]], i1 noundef false, i1 noundef true, i1 noundef true)
 ; CHECK-NEXT:    call void @unknown_no_openmp()
 ; CHECK-NEXT:    [[TMP0:%.*]] = icmp eq i32 [[CALL]], -1
 ; CHECK-NEXT:    br i1 [[TMP0]], label [[MASTER:%.*]], label [[EXIT:%.*]]
 ; CHECK:       master:
-; CHECK-NEXT:    call void @use.internalized(i8* nofree writeonly addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @y, i32 0, i32 0) to i8*)) #[[ATTR4]]
+; CHECK-NEXT:    call void @use.internalized(i8* nofree writeonly addrspacecast (i8 addrspace(3)* getelementptr inbounds ([4 x i8], [4 x i8] addrspace(3)* @y, i32 0, i32 0) to i8*)) #[[ATTR5]]
 ; CHECK-NEXT:    br label [[EXIT]]
 ; CHECK:       exit:
 ; CHECK-NEXT:    ret void
 ;
 ;
 ; CHECK-LABEL: define {{[^@]+}}@negative_qux_spmd() {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[CALL:%.*]] = call i32 @__kmpc_target_init(%struct.ident_t* noundef nonnull @[[GLOB1]], i1 noundef true, i1 noundef true, i1 noundef true)
 ; CHECK-NEXT:    call void @unknown_no_openmp()
 ; CHECK-NEXT:    [[TMP0:%.*]] = icmp eq i32 [[CALL]], -1
 ; CHECK-NEXT:    br i1 [[TMP0]], label [[MASTER:%.*]], label [[EXIT:%.*]]
 ; CHECK:       master:
-; CHECK-NEXT:    [[Y:%.*]] = call i8* @__kmpc_alloc_shared(i64 noundef 24) #[[ATTR1]], !dbg [[DBG9:![0-9]+]]
-; CHECK-NEXT:    call void @use.internalized(i8* nofree writeonly [[Y]]) #[[ATTR4]]
-; CHECK-NEXT:    call void @__kmpc_free_shared(i8* [[Y]], i64 noundef 24) #[[ATTR1]]
+; CHECK-NEXT:    [[Y:%.*]] = call i8* @__kmpc_alloc_shared(i64 noundef 24) #[[ATTR4]], !dbg [[DBG9:![0-9]+]]
+; CHECK-NEXT:    call void @use.internalized(i8* nofree writeonly [[Y]]) #[[ATTR5]]
+; CHECK-NEXT:    call void @__kmpc_free_shared(i8* [[Y]], i64 noundef 24) #[[ATTR4]]
 ; CHECK-NEXT:    br label [[EXIT]]
 ; CHECK:       exit:
 ; CHECK-NEXT:    ret void
 ;
 ;
+; CHECK: Function Attrs: nofree nosync nounwind willreturn writeonly
 ; CHECK-LABEL: define {{[^@]+}}@use.internalized
 ; CHECK-SAME: (i8* nofree writeonly [[X:%.*]]) #[[ATTR0:[0-9]+]] {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    store i8* [[X]], i8** @S, align 8
 ; CHECK-NEXT:    ret void
 ;
 ;
 ; CHECK-LABEL: define {{[^@]+}}@use
 ; CHECK-SAME: (i8* [[X:%.*]]) {
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    store i8* [[X]], i8** @S, align 8
 ; CHECK-NEXT:    ret void
 ;
 ;.
 ; CHECK: attributes #[[ATTR0]] = { nofree nosync nounwind willreturn writeonly }
-; CHECK: attributes #[[ATTR1]] = { nounwind }
+; CHECK: attributes #[[ATTR1:[0-9]+]] = { nosync nounwind }
 ; CHECK: attributes #[[ATTR2:[0-9]+]] = { nounwind readnone }
 ; CHECK: attributes #[[ATTR3:[0-9]+]] = { "llvm.assume"="omp_no_openmp" }
-; CHECK: attributes #[[ATTR4]] = { nounwind writeonly }
+; CHECK: attributes #[[ATTR4]] = { nounwind }
+; CHECK: attributes #[[ATTR5]] = { nounwind writeonly }
 ;.
 ; CHECK: [[META0:![0-9]+]] = distinct !DICompileUnit(language: DW_LANG_C99, file: !1, producer: "clang version 12.0.0", isOptimized: false, runtimeVersion: 0, emissionKind: FullDebug, enums: !2, splitDebugInlining: false, nameTableKind: None)
 ; CHECK: [[META1:![0-9]+]] = !DIFile(filename: "replace_globalization.c", directory: "/tmp/replace_globalization.c")
 ; CHECK: [[META2:![0-9]+]] = !{}
 ; CHECK: [[META3:![0-9]+]] = !{i32 2, !"Debug Info Version", i32 3}
 ; CHECK: [[META4:![0-9]+]] = !{i32 1, !"wchar_size", i32 4}
 ; CHECK: [[META5:![0-9]+]] = !{i32 7, !"openmp", i32 50}
 ; CHECK: [[META6:![0-9]+]] = !{i32 7, !"openmp-device", i32 50}
 ; CHECK: [[META7:![0-9]+]] = !{void ()* @foo, !"kernel", i32 1}
 ; CHECK: [[META8:![0-9]+]] = !{void ()* @bar, !"kernel", i32 1}
 ; CHECK: [[DBG9]] = !DILocation(line: 5, column: 14, scope: !10)
 ; CHECK: [[META10:![0-9]+]] = distinct !DISubprogram(name: "bar", scope: !1, file: !1, line: 1, type: !11, scopeLine: 1, flags: DIFlagPrototyped, spFlags: DISPFlagDefinition | DISPFlagOptimized, unit: !0, retainedNodes: !2)
 ; CHECK: [[META11:![0-9]+]] = !DISubroutineType(types: !2)
 ;.