Fix VLA unnecessary allocation. See more

- Emit element type in vla DSA to fix allocation size in lowering
- Skip shared vlas in additional space allocation

Fixes llvm#180

Author: Raúl Peñacoba Veigas

clang/lib/CodeGen/CGOmpSsRuntime.cpp

@@ -1188,13 +1188,9 @@ void CGOmpSsRuntime::EmitDSAShared(
       CaptureMapStack.back().try_emplace(VD, LV.getAddress(CGF));
       DSAValue = LV.getPointer(CGF);
       DSABundleList.push_back(DSAValue);
-      DSABundleList.push_back(llvm::UndefValue::get(CGF.ConvertType(DRE->getType())));
-      TaskInfo.emplace_back(getBundleStr(OSSB_shared), DSABundleList);
     } else {
       DSAValue = CGF.EmitDeclRefLValue(DRE).getPointer(CGF);
       DSABundleList.push_back(DSAValue);
-      DSABundleList.push_back(llvm::UndefValue::get(CGF.ConvertType(DRE->getType())));
-      TaskInfo.emplace_back(getBundleStr(OSSB_shared), DSABundleList);
     }
     QualType Q = VD->getType();
     // int (**p)[sizex][sizey] -> we need to capture sizex sizey only
@@ -1203,8 +1199,15 @@ void CGOmpSsRuntime::EmitDSAShared(
     while (Q->isPointerType()) {
       Q = Q->getPointeeType();
     }
-    if (Q->isVariableArrayType())
+    if (Q->isVariableArrayType()) {
       GatherVLADims(CGF, DSAValue, Q, DimsWithValue, CapturedList, IsPtr);
+      QualType BaseElementTy = CGF.getContext().getBaseElementType(DRE->getType());
+      DSABundleList.push_back(llvm::UndefValue::get(CGF.ConvertType(BaseElementTy)));
+    } else {
+      DSABundleList.push_back(llvm::UndefValue::get(CGF.ConvertType(DRE->getType())));
+    }
+
+    TaskInfo.emplace_back(getBundleStr(OSSB_shared), DSABundleList);
 
     if (!DimsWithValue.empty())
       TaskInfo.emplace_back(getBundleStr(OSSB_vladims), DimsWithValue);
@@ -1237,13 +1240,9 @@ void CGOmpSsRuntime::EmitDSAPrivate(
     CaptureMapStack.back().try_emplace(VD, LV.getAddress(CGF));
     DSAValue = LV.getPointer(CGF);
     DSABundleList.push_back(DSAValue);
-    DSABundleList.push_back(llvm::UndefValue::get(CGF.ConvertType(DRE->getType())));
-    TaskInfo.emplace_back(getBundleStr(OSSB_private), DSABundleList);
   } else {
     DSAValue = CGF.EmitDeclRefLValue(DRE).getPointer(CGF);
     DSABundleList.push_back(DSAValue);
-    DSABundleList.push_back(llvm::UndefValue::get(CGF.ConvertType(DRE->getType())));
-    TaskInfo.emplace_back(getBundleStr(OSSB_private), DSABundleList);
   }
   QualType Q = VD->getType();
   // int (**p)[sizex][sizey] -> we need to capture sizex sizey only
@@ -1252,8 +1251,15 @@ void CGOmpSsRuntime::EmitDSAPrivate(
   while (Q->isPointerType()) {
     Q = Q->getPointeeType();
   }
-  if (Q->isVariableArrayType())
+  if (Q->isVariableArrayType()) {
     GatherVLADims(CGF, DSAValue, Q, DimsWithValue, CapturedList, IsPtr);
+    QualType BaseElementTy = CGF.getContext().getBaseElementType(DRE->getType());
+    DSABundleList.push_back(llvm::UndefValue::get(CGF.ConvertType(BaseElementTy)));
+  } else {
+    DSABundleList.push_back(llvm::UndefValue::get(CGF.ConvertType(DRE->getType())));
+  }
+
+  TaskInfo.emplace_back(getBundleStr(OSSB_private), DSABundleList);
 
   if (!DimsWithValue.empty())
     TaskInfo.emplace_back(getBundleStr(OSSB_vladims), DimsWithValue);
@@ -1291,13 +1297,9 @@ void CGOmpSsRuntime::EmitDSAFirstprivate(
     CaptureMapStack.back().try_emplace(VD, LV.getAddress(CGF));
     DSAValue = LV.getPointer(CGF);
     DSABundleList.push_back(DSAValue);
-    DSABundleList.push_back(llvm::UndefValue::get(CGF.ConvertType(DRE->getType())));
-    TaskInfo.emplace_back(getBundleStr(OSSB_firstprivate), DSABundleList);
   } else {
     DSAValue = CGF.EmitDeclRefLValue(DRE).getPointer(CGF);
     DSABundleList.push_back(DSAValue);
-    DSABundleList.push_back(llvm::UndefValue::get(CGF.ConvertType(DRE->getType())));
-    TaskInfo.emplace_back(getBundleStr(OSSB_firstprivate), DSABundleList);
   }
   QualType Q = VD->getType();
   // int (**p)[sizex][sizey] -> we need to capture sizex sizey only
@@ -1306,8 +1308,15 @@ void CGOmpSsRuntime::EmitDSAFirstprivate(
   while (Q->isPointerType()) {
     Q = Q->getPointeeType();
   }
-  if (Q->isVariableArrayType())
+  if (Q->isVariableArrayType()) {
     GatherVLADims(CGF, DSAValue, Q, DimsWithValue, CapturedList, IsPtr);
+    QualType BaseElementTy = CGF.getContext().getBaseElementType(DRE->getType());
+    DSABundleList.push_back(llvm::UndefValue::get(CGF.ConvertType(BaseElementTy)));
+  } else {
+    DSABundleList.push_back(llvm::UndefValue::get(CGF.ConvertType(DRE->getType())));
+  }
+
+  TaskInfo.emplace_back(getBundleStr(OSSB_firstprivate), DSABundleList);
 
   if (!DimsWithValue.empty())
     TaskInfo.emplace_back(getBundleStr(OSSB_vladims), DimsWithValue);

clang/test/OmpSs/IR/task_depend_array_section.c

@@ -488,6 +488,10 @@ void DirectiveEnvironment::verifyVLADimsInfo() {
   for (const auto &VLAWithDimsMap : VLADimsInfo) {
     if (!valueInDSABundles(VLAWithDimsMap.first))
       llvm_unreachable("VLA dims OperandBundle must have an associated DSA");
+    if (!(getDSAType(VLAWithDimsMap.first)->isSingleValueType()
+          || getDSAType(VLAWithDimsMap.first)->isStructTy()))
+      llvm_unreachable("VLA type is not scalar");
+
     // VLA Dims that are not Captured is an error
     for (auto *V : VLAWithDimsMap.second) {
       if (!valueInCapturedBundle(V))

clang/test/OmpSs/IR/task_vla.c

@@ -1480,6 +1480,9 @@ struct OmpSsDirective {
   Value *computeTaskArgsVLAsExtraSizeOf(IRBuilder<> &IRB) {
     Value *Sum = ConstantInt::get(Int64Ty, 0);
     for (const auto &VLAWithDimsMap : VLADimsInfo) {
+      // Skip shareds because they don't need space in task_args
+      if (DSAInfo.Shared.count(VLAWithDimsMap.first))
+        continue;
       Type *Ty = DirEnv.getDSAType(VLAWithDimsMap.first);
       unsigned SizeB = DL.getTypeAllocSize(Ty);
       Value *ArraySize = ConstantInt::get(Int64Ty, SizeB);
@@ -1636,6 +1639,9 @@ struct OmpSsDirective {
   // Greater alignemt go first
   void computeVLAsAlignOrder(SmallVectorImpl<VLAAlign> &VLAAlignsInfo) {
     for (const auto &VLAWithDimsMap : VLADimsInfo) {
+      // Skip shareds because they don't need space in task_args
+      if (DSAInfo.Shared.count(VLAWithDimsMap.first))
+        continue;
       auto *V = VLAWithDimsMap.first;
       Type *Ty = DirEnv.getDSAType(V);
 

llvm/lib/Analysis/OmpSsRegionAnalysis.cpp

@@ -0,0 +1,176 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --include-generated-funcs --version 2
+; RUN: opt %s -passes='default<O3>,ompss-2' -S | FileCheck %s
+; ModuleID = 'check_vla_size_in_task_args.ll'
+source_filename = "check_vla_size_in_task_args.ll"
+target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+; void foo() {
+;   int n = 1;
+;   int vla[n][20];
+;   #pragma oss task firstprivate(vla)
+;   {}
+; }
+
+; Function Attrs: noinline nounwind
+define dso_local void @foo() #0 !dbg !5 {
+entry:
+  %n = alloca i32, align 4
+  %saved_stack = alloca ptr, align 8
+  %__vla_expr0 = alloca i64, align 8
+  store i32 1, ptr %n, align 4, !dbg !9
+  %0 = load i32, ptr %n, align 4, !dbg !10
+  %1 = zext i32 %0 to i64, !dbg !11
+  %2 = call ptr @llvm.stacksave(), !dbg !11
+  store ptr %2, ptr %saved_stack, align 8, !dbg !11
+  %vla = alloca [20 x i32], i64 %1, align 16, !dbg !11
+  store i64 %1, ptr %__vla_expr0, align 8, !dbg !11
+  %3 = call token @llvm.directive.region.entry() [ "DIR.OSS"([5 x i8] c"TASK\00"), "QUAL.OSS.FIRSTPRIVATE"(ptr %vla, i32 undef), "QUAL.OSS.VLA.DIMS"(ptr %vla, i64 %1, i64 20), "QUAL.OSS.CAPTURED"(i64 %1, i64 20) ], !dbg !12
+  call void @llvm.directive.region.exit(token %3), !dbg !13
+  %4 = load ptr, ptr %saved_stack, align 8, !dbg !14
+  call void @llvm.stackrestore(ptr %4), !dbg !14
+  ret void, !dbg !14
+}
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn
+declare ptr @llvm.stacksave() #1
+
+; Function Attrs: nounwind
+declare token @llvm.directive.region.entry() #2
+
+; Function Attrs: nounwind
+declare void @llvm.directive.region.exit(token) #2
+
+; Function Attrs: nocallback nofree nosync nounwind willreturn
+declare void @llvm.stackrestore(ptr) #1
+
+attributes #0 = { noinline nounwind "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" "target-features"="+cx8,+mmx,+sse,+sse2,+x87" }
+attributes #1 = { nocallback nofree nosync nounwind willreturn }
+attributes #2 = { nounwind }
+
+!llvm.dbg.cu = !{!0}
+!llvm.module.flags = !{!2, !3}
+!llvm.ident = !{!4}
+
+!0 = distinct !DICompileUnit(language: DW_LANG_C11, file: !1, producer: "clang version 17.0.0 ([email protected]:llvm-ompss/llvm-mono.git 189d6f1b886a40340f7dd586e6beeca3482d3f5b)", isOptimized: false, runtimeVersion: 0, emissionKind: NoDebug, splitDebugInlining: false, nameTableKind: None)
+!1 = !DIFile(filename: "<stdin>", directory: "")
+!2 = !{i32 2, !"Debug Info Version", i32 3}
+!3 = !{i32 1, !"wchar_size", i32 4}
+!4 = !{!"clang version 17.0.0 ([email protected]:llvm-ompss/llvm-mono.git 189d6f1b886a40340f7dd586e6beeca3482d3f5b)"}
+!5 = distinct !DISubprogram(name: "foo", scope: !6, file: !6, line: 1, type: !7, scopeLine: 1, spFlags: DISPFlagDefinition, unit: !0, retainedNodes: !8)
+!6 = !DIFile(filename: "check_vla_size_in_task_args.ll", directory: "")
+!7 = !DISubroutineType(types: !8)
+!8 = !{}
+!9 = !DILocation(line: 2, column: 7, scope: !5)
+!10 = !DILocation(line: 3, column: 11, scope: !5)
+!11 = !DILocation(line: 3, column: 3, scope: !5)
+!12 = !DILocation(line: 4, column: 11, scope: !5)
+!13 = !DILocation(line: 5, column: 4, scope: !5)
+!14 = !DILocation(line: 6, column: 1, scope: !5)
+; CHECK-LABEL: define dso_local void @foo
+; CHECK-SAME: () local_unnamed_addr #[[ATTR0:[0-9]+]] !dbg [[DBG5:![0-9]+]] {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[VLA:%.*]] = alloca [20 x i32], align 16, !dbg [[DBG9:![0-9]+]]
+; CHECK-NEXT:    [[TMP0:%.*]] = alloca ptr, align 8, !dbg [[DBG10:![0-9]+]]
+; CHECK-NEXT:    [[TMP1:%.*]] = alloca ptr, align 8, !dbg [[DBG10]]
+; CHECK-NEXT:    [[NUM_DEPS:%.*]] = alloca i64, align 8, !dbg [[DBG10]]
+; CHECK-NEXT:    br label [[FINAL_COND:%.*]], !dbg [[DBG10]]
+; CHECK:       codeRepl:
+; CHECK-NEXT:    store i64 0, ptr [[NUM_DEPS]], align 8, !dbg [[DBG10]]
+; CHECK-NEXT:    [[TMP2:%.*]] = load i64, ptr [[NUM_DEPS]], align 8, !dbg [[DBG10]]
+; CHECK-NEXT:    call void @nanos6_create_task(ptr @task_info_var_foo, ptr @task_invocation_info_foo, ptr null, i64 112, ptr [[TMP0]], ptr [[TMP1]], i64 0, i64 [[TMP2]]), !dbg [[DBG10]]
+; CHECK-NEXT:    [[TMP3:%.*]] = load ptr, ptr [[TMP0]], align 8, !dbg [[DBG10]]
+; CHECK-NEXT:    [[ARGS_END:%.*]] = getelementptr i8, ptr [[TMP3]], i64 32, !dbg [[DBG10]]
+; CHECK-NEXT:    [[GEP_VLA:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO:%.*]], ptr [[TMP3]], i32 0, i32 0, !dbg [[DBG10]]
+; CHECK-NEXT:    store ptr [[ARGS_END]], ptr [[GEP_VLA]], align 4, !dbg [[DBG10]]
+; CHECK-NEXT:    [[TMP4:%.*]] = getelementptr i8, ptr [[ARGS_END]], i64 80, !dbg [[DBG10]]
+; CHECK-NEXT:    [[GEP_VLA1:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TMP3]], i32 0, i32 0, !dbg [[DBG10]]
+; CHECK-NEXT:    [[TMP5:%.*]] = load ptr, ptr [[GEP_VLA1]], align 8, !dbg [[DBG10]]
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[TMP5]], ptr align 4 [[VLA]], i64 80, i1 false), !dbg [[DBG10]]
+; CHECK-NEXT:    [[CAPT_GEP_:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TMP3]], i32 0, i32 1, !dbg [[DBG10]]
+; CHECK-NEXT:    store i64 1, ptr [[CAPT_GEP_]], align 8, !dbg [[DBG10]]
+; CHECK-NEXT:    [[CAPT_GEP_2:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TMP3]], i32 0, i32 2, !dbg [[DBG10]]
+; CHECK-NEXT:    store i64 20, ptr [[CAPT_GEP_2]], align 8, !dbg [[DBG10]]
+; CHECK-NEXT:    [[TMP6:%.*]] = load ptr, ptr [[TMP1]], align 8, !dbg [[DBG10]]
+; CHECK-NEXT:    call void @nanos6_submit_task(ptr [[TMP6]]), !dbg [[DBG10]]
+; CHECK-NEXT:    br label [[FINAL_END:%.*]], !dbg [[DBG10]]
+; CHECK:       final.end:
+; CHECK-NEXT:    ret void, !dbg [[DBG11:![0-9]+]]
+; CHECK:       final.then:
+; CHECK-NEXT:    br label [[FINAL_END]], !dbg [[DBG11]]
+; CHECK:       final.cond:
+; CHECK-NEXT:    [[TMP7:%.*]] = call i32 @nanos6_in_final(), !dbg [[DBG10]]
+; CHECK-NEXT:    [[TMP8:%.*]] = icmp ne i32 [[TMP7]], 0, !dbg [[DBG10]]
+; CHECK-NEXT:    br i1 [[TMP8]], label [[FINAL_THEN:%.*]], label [[CODEREPL:%.*]], !dbg [[DBG10]]
+;
+;
+; CHECK-LABEL: define internal void @nanos6_ol_duplicate_foo
+; CHECK-SAME: (ptr [[TASK_ARGS_SRC:%.*]], ptr [[TASK_ARGS_DST:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = load ptr, ptr [[TASK_ARGS_DST]], align 8
+; CHECK-NEXT:    [[ARGS_END:%.*]] = getelementptr i8, ptr [[TMP0]], i64 32
+; CHECK-NEXT:    [[GEP_DST_VLA:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO:%.*]], ptr [[TMP0]], i32 0, i32 0
+; CHECK-NEXT:    store ptr [[ARGS_END]], ptr [[GEP_DST_VLA]], align 4
+; CHECK-NEXT:    [[GEP_DST_:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TMP0]], i32 0, i32 1
+; CHECK-NEXT:    [[TMP1:%.*]] = load i64, ptr [[GEP_DST_]], align 8
+; CHECK-NEXT:    [[TMP2:%.*]] = mul nuw i64 4, [[TMP1]]
+; CHECK-NEXT:    [[GEP_DST_1:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TMP0]], i32 0, i32 2
+; CHECK-NEXT:    [[TMP3:%.*]] = load i64, ptr [[GEP_DST_1]], align 8
+; CHECK-NEXT:    [[TMP4:%.*]] = mul nuw i64 [[TMP2]], [[TMP3]]
+; CHECK-NEXT:    [[TMP5:%.*]] = getelementptr i8, ptr [[ARGS_END]], i64 [[TMP4]]
+; CHECK-NEXT:    [[GEP_SRC_:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TASK_ARGS_SRC]], i32 0, i32 1
+; CHECK-NEXT:    [[TMP6:%.*]] = load i64, ptr [[GEP_SRC_]], align 8
+; CHECK-NEXT:    [[TMP7:%.*]] = mul nuw i64 1, [[TMP6]]
+; CHECK-NEXT:    [[GEP_SRC_2:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TASK_ARGS_SRC]], i32 0, i32 2
+; CHECK-NEXT:    [[TMP8:%.*]] = load i64, ptr [[GEP_SRC_2]], align 8
+; CHECK-NEXT:    [[TMP9:%.*]] = mul nuw i64 [[TMP7]], [[TMP8]]
+; CHECK-NEXT:    [[GEP_SRC_VLA:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TASK_ARGS_SRC]], i32 0, i32 0
+; CHECK-NEXT:    [[GEP_DST_VLA3:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TMP0]], i32 0, i32 0
+; CHECK-NEXT:    [[TMP10:%.*]] = load ptr, ptr [[GEP_SRC_VLA]], align 8
+; CHECK-NEXT:    [[TMP11:%.*]] = load ptr, ptr [[GEP_DST_VLA3]], align 8
+; CHECK-NEXT:    [[TMP12:%.*]] = mul nuw i64 [[TMP9]], 4
+; CHECK-NEXT:    call void @llvm.memcpy.p0.p0.i64(ptr align 4 [[TMP11]], ptr align 4 [[TMP10]], i64 [[TMP12]], i1 false)
+; CHECK-NEXT:    [[CAPT_GEP_SRC_:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TASK_ARGS_SRC]], i32 0, i32 1
+; CHECK-NEXT:    [[CAPT_GEP_DST_:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TMP0]], i32 0, i32 1
+; CHECK-NEXT:    [[TMP13:%.*]] = load i64, ptr [[CAPT_GEP_SRC_]], align 8
+; CHECK-NEXT:    store i64 [[TMP13]], ptr [[CAPT_GEP_DST_]], align 8
+; CHECK-NEXT:    [[CAPT_GEP_SRC_4:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TASK_ARGS_SRC]], i32 0, i32 2
+; CHECK-NEXT:    [[CAPT_GEP_DST_5:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TMP0]], i32 0, i32 2
+; CHECK-NEXT:    [[TMP14:%.*]] = load i64, ptr [[CAPT_GEP_SRC_4]], align 8
+; CHECK-NEXT:    store i64 [[TMP14]], ptr [[CAPT_GEP_DST_5]], align 8
+; CHECK-NEXT:    ret void
+;
+;
+; CHECK-LABEL: define internal void @nanos6_unpacked_task_region_foo
+; CHECK-SAME: (ptr [[VLA:%.*]], i64 [[TMP0:%.*]], i64 [[TMP1:%.*]], ptr [[DEVICE_ENV:%.*]], ptr [[ADDRESS_TRANSLATION_TABLE:%.*]]) !dbg [[DBG12:![0-9]+]] {
+; CHECK-NEXT:  newFuncRoot:
+; CHECK-NEXT:    br label [[TMP2:%.*]], !dbg [[DBG13:![0-9]+]]
+; CHECK:       2:
+; CHECK-NEXT:    br label [[DOTEXITSTUB:%.*]], !dbg [[DBG14:![0-9]+]]
+; CHECK:       .exitStub:
+; CHECK-NEXT:    ret void
+;
+;
+; CHECK-LABEL: define internal void @nanos6_ol_task_region_foo
+; CHECK-SAME: (ptr [[TASK_ARGS:%.*]], ptr [[DEVICE_ENV:%.*]], ptr [[ADDRESS_TRANSLATION_TABLE:%.*]]) {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[GEP_VLA:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO:%.*]], ptr [[TASK_ARGS]], i32 0, i32 0
+; CHECK-NEXT:    [[LOAD_GEP_VLA:%.*]] = load ptr, ptr [[GEP_VLA]], align 8
+; CHECK-NEXT:    [[CAPT_GEP:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TASK_ARGS]], i32 0, i32 1
+; CHECK-NEXT:    [[LOAD_CAPT_GEP:%.*]] = load i64, ptr [[CAPT_GEP]], align 8
+; CHECK-NEXT:    [[CAPT_GEP1:%.*]] = getelementptr [[NANOS6_TASK_ARGS_FOO]], ptr [[TASK_ARGS]], i32 0, i32 2
+; CHECK-NEXT:    [[LOAD_CAPT_GEP1:%.*]] = load i64, ptr [[CAPT_GEP1]], align 8
+; CHECK-NEXT:    [[TMP0:%.*]] = icmp ne ptr [[ADDRESS_TRANSLATION_TABLE]], null
+; CHECK-NEXT:    br i1 [[TMP0]], label [[TMP1:%.*]], label [[TMP2:%.*]]
+; CHECK:       1:
+; CHECK-NEXT:    br label [[TMP2]]
+; CHECK:       2:
+; CHECK-NEXT:    call void @nanos6_unpacked_task_region_foo(ptr [[LOAD_GEP_VLA]], i64 [[LOAD_CAPT_GEP]], i64 [[LOAD_CAPT_GEP1]], ptr [[DEVICE_ENV]], ptr [[ADDRESS_TRANSLATION_TABLE]])
+; CHECK-NEXT:    ret void
+;
+;
+; CHECK-LABEL: define internal void @nanos6_constructor_register_task_info() {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    call void @nanos6_register_task_info(ptr @task_info_var_foo)
+; CHECK-NEXT:    ret void
+;