[HLSL] [DirectX backend] Move generateGlobalCtorDtorCalls into DirectX backend.

[python3kgae]

Summary

Move generateGlobalCtorDtorCalls out of clang CodeGen.
A new transform pass GlobalCtorDtorCalls is created to do the job.
GlobalCtorDtorCalls is registed in registerPipelineEarlySimplificationEPCallback.

[efriedma]

You might actually want to run this twice, once early, and once in the backend. Early to get high-quality optimization, late in case optimizations don't run for some reason. (Not sure how that works if T.getEnvironment() == Triple::EnvironmentType::Library, though, since we don't actually erase the constructors?)

[python3kgae]

In D135429#3843906, @efriedma wrote:

You might actually want to run this twice, once early, and once in the backend. Early to get high-quality optimization, late in case optimizations don't run for some reason. (Not sure how that works if T.getEnvironment() == Triple::EnvironmentType::Library, though, since we don't actually erase the constructors?)

The pass will be always enabled in registerPipelineEarlySimplificationEPCallback (Not controlled by optimization level), is it possible that things registered in registerPipelineEarlySimplificationEPCallback be skipped?

For the library case, we'll have to erase the constructors after generating calls for them in the entry function when linking.
In case the target profile is still library when linking, we'll have to merge the constructors.
These should be done in the linker. Created https://github.com/llvm/llvm-project/issues/58238 to track it.

[efriedma]

The pass will be always enabled in registerPipelineEarlySimplificationEPCallback (Not controlled by optimization level), is it possible that things registered in registerPipelineEarlySimplificationEPCallback be skipped?

I think we won't end up running it at -O0.

[efriedma]

In D135429#3844172, @efriedma wrote:

The pass will be always enabled in registerPipelineEarlySimplificationEPCallback (Not controlled by optimization level), is it possible that things registered in registerPipelineEarlySimplificationEPCallback be skipped?

I think we won't end up running it at -O0.

Oh, hmm, just checked, and it looks like we have a special codepath specifically to ensure PipelineEarlySimplificationEPCallbacks runs at -O0. I guess you're fine in that respect.

[python3kgae]

In D135429#3844179, @efriedma wrote:

In D135429#3844172, @efriedma wrote:

The pass will be always enabled in registerPipelineEarlySimplificationEPCallback (Not controlled by optimization level), is it possible that things registered in registerPipelineEarlySimplificationEPCallback be skipped?

I think we won't end up running it at -O0.

Oh, hmm, just checked, and it looks like we have a special codepath specifically to ensure PipelineEarlySimplificationEPCallbacks runs at -O0. I guess you're fine in that respect.

I see. Thanks for the update.

[python3kgae]

Add end of file newline

[python3kgae]

gentle ping

[beanz]

I don’t think you should be deleting all those frontend tests. Those tests existed before your change to remove the global constructor global variable, they provide valuable coverage of the frontend constructor generation.

[beanz]

In addition to restoring the test cases that you deleted, I think you should also reduce and simplify the test cases you're adding. These new tests are very large blobs of IR generated by clang. They include a lot of extra instructions that aren't needed to exercise the code you wrote in this patch.

[beanz]

clang/lib/CodeGen/CGHLSLRuntime.cpp

193

I think you've been a bit too aggressive about what code you're moving. The code that generates the global constructor calls should not be removed.

We should generate correct IR from clang, and part of correct IR for HLSL is that the global constructors are called from the entries.

clang/lib/CodeGen/CGHLSLRuntime.h

protected:
 
   llvm::Value *emitInputSemantic(llvm::IRBuilder<> &B, const ParmVarDecl &D);
 
 public:
   CGHLSLRuntime(CodeGenModule &CGM) : CGM(CGM) {}
   virtual ~CGHLSLRuntime() {}
 
   void annotateHLSLResource(const VarDecl *D, llvm::GlobalVariable *GV);
-  void generateGlobalCtorDtorCalls();
 
   void finishCodeGen();
 
   void setHLSLEntryAttributes(const FunctionDecl *FD, llvm::Function *Fn);
 
   void emitEntryFunction(const FunctionDecl *FD, llvm::Function *Fn);
 };
 
 } // namespace CodeGen
 } // namespace clang
 
 #endif

clang/lib/CodeGen/CGHLSLRuntime.cpp

 
 void CGHLSLRuntime::finishCodeGen() {
   auto &TargetOpts = CGM.getTarget().getTargetOpts();
   llvm::Module &M = CGM.getModule();
   Triple T(M.getTargetTriple());
   if (T.getArch() == Triple::ArchType::dxil)
     addDxilValVersion(TargetOpts.DxilValidatorVersion, M);
 
-  generateGlobalCtorDtorCalls();
   if (CGM.getCodeGenOpts().OptimizationLevel == 0)
     addDisableOptimizations(M);
 }
 
 void CGHLSLRuntime::annotateHLSLResource(const VarDecl *D, GlobalVariable *GV) {
   const Type *Ty = D->getType()->getPointeeOrArrayElementType();
   if (!Ty)
     return;
void CGHLSLRuntime::emitEntryFunction(const FunctionDecl *FD,
   }
 
   CallInst *CI = B.CreateCall(FunctionCallee(Fn), Args);
   (void)CI;
   // FIXME: Handle codegen for return type semantics.
   // See: https://github.com/llvm/llvm-project/issues/57875
   B.CreateRetVoid();
 }
-
-static void gatherFunctions(SmallVectorImpl<Function *> &Fns, llvm::Module &M,
-                            bool CtorOrDtor) {
-  const auto *GV =
-      M.getNamedGlobal(CtorOrDtor ? "llvm.global_ctors" : "llvm.global_dtors");
-  if (!GV)
-    return;
-  const auto *CA = dyn_cast<ConstantArray>(GV->getInitializer());
-  if (!CA)
-    return;
-  // The global_ctor array elements are a struct [Priority, Fn *, COMDat].
-  // HLSL neither supports priorities or COMDat values, so we will check those
-  // in an assert but not handle them.
-
-  llvm::SmallVector<Function *> CtorFns;
-  for (const auto &Ctor : CA->operands()) {
-    if (isa<ConstantAggregateZero>(Ctor))
-      continue;
-    ConstantStruct *CS = cast<ConstantStruct>(Ctor);
-
-    assert(cast<ConstantInt>(CS->getOperand(0))->getValue() == 65535 &&
-           "HLSL doesn't support setting priority for global ctors.");
-    assert(isa<ConstantPointerNull>(CS->getOperand(2)) &&
-           "HLSL doesn't support COMDat for global ctors.");
-    Fns.push_back(cast<Function>(CS->getOperand(1)));
-  }
-}
-
-void CGHLSLRuntime::generateGlobalCtorDtorCalls() {
-  llvm::Module &M = CGM.getModule();
-  SmallVector<Function *> CtorFns;
-  SmallVector<Function *> DtorFns;
-  gatherFunctions(CtorFns, M, true);
-  gatherFunctions(DtorFns, M, false);
-
-  // Insert a call to the global constructor at the beginning of the entry block
-  // to externally exported functions. This is a bit of a hack, but HLSL allows
-  // global constructors, but doesn't support driver initialization of globals.
-  for (auto &F : M.functions()) {
-    if (!F.hasFnAttribute("hlsl.shader"))
-      continue;
-    IRBuilder<> B(&F.getEntryBlock(), F.getEntryBlock().begin());
-    for (auto *Fn : CtorFns)
-      B.CreateCall(FunctionCallee(Fn));
-
-    // Insert global dtors before the terminator of the last instruction
-    B.SetInsertPoint(F.back().getTerminator());
-    for (auto *Fn : DtorFns)
-      B.CreateCall(FunctionCallee(Fn));
-  }
-
-  // No need to keep global ctors/dtors for non-lib profile after call to
-  // ctors/dtors added for entry.
-  Triple T(M.getTargetTriple());
-  if (T.getEnvironment() != Triple::EnvironmentType::Library) {
-    if (auto *GV = M.getNamedGlobal("llvm.global_ctors"))
-      GV->eraseFromParent();
-    if (auto *GV = M.getNamedGlobal("llvm.global_dtors"))
-      GV->eraseFromParent();
-  }
-}

clang/test/CodeGenHLSL/GlobalConstructorFunction.hlsl

-// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.0-compute -S -emit-llvm -disable-llvm-passes %s -o - | FileCheck %s
-
-int i;
-
-__attribute__((constructor)) void call_me_first(void) {
-  i = 12;
-}
-
-__attribute__((constructor)) void then_call_me(void) {
-  i = 12;
-}
-
-__attribute__((destructor)) void call_me_last(void) {
-  i = 0;
-}
-
-[numthreads(1,1,1)]
-void main(unsigned GI : SV_GroupIndex) {}
-
-// Make sure global variable for ctors/dtors removed.
-// CHECK-NOT:@llvm.global_ctors
-// CHECK-NOT:@llvm.global_dtors
-
-//CHECK: define void @main()
-//CHECK-NEXT: entry:
-//CHECK-NEXT:   call void @"?call_me_first@@YAXXZ"()
-//CHECK-NEXT:   call void @"?then_call_me@@YAXXZ"()
-//CHECK-NEXT:   %0 = call i32 @llvm.dx.flattened.thread.id.in.group()
-//CHECK-NEXT:   call void @"?main@@YAXI@Z"(i32 %0)
-//CHECK-NEXT:   call void @"?call_me_last@@YAXXZ"(
-//CHECK-NEXT:   ret void

clang/test/CodeGenHLSL/GlobalConstructorLib.hlsl

-// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.3-library -S -emit-llvm -disable-llvm-passes %s -o - | FileCheck %s
-
-// Make sure global variable for ctors exist for lib profile.
-// CHECK:@llvm.global_ctors
-
-RWBuffer<float> Buffer;
-
-[shader("compute")]
-[numthreads(1,1,1)]
-void FirstEntry() {}
-
-// CHECK: define void @FirstEntry()
-// CHECK-NEXT: entry:
-// CHECK-NEXT:   call void @_GLOBAL__sub_I_GlobalConstructorLib.hlsl()
-
-[shader("compute")]
-[numthreads(1,1,1)]
-void SecondEntry() {}
-
-// CHECK: define void @SecondEntry()
-// CHECK-NEXT: entry:
-// CHECK-NEXT:   call void @_GLOBAL__sub_I_GlobalConstructorLib.hlsl()
-// CHECK-NEXT:   call void @"?SecondEntry@@YAXXZ"()

clang/test/CodeGenHLSL/GlobalConstructors.hlsl

-// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.0-compute -x hlsl -S -emit-llvm -disable-llvm-passes %s -o - | FileCheck %s
-
-RWBuffer<float> Buffer;
-
-[numthreads(1,1,1)]
-void main(unsigned GI : SV_GroupIndex) {}
-
-// Make sure global variable for ctors/dtors removed.
-// CHECK-NOT:@llvm.global_ctors
-// CHECK-NOT:@llvm.global_dtors
-//CHECK:      define void @main()
-//CHECK-NEXT: entry:
-//CHECK-NEXT:   call void @_GLOBAL__sub_I_GlobalConstructors.hlsl()
-//CHECK-NEXT:   %0 = call i32 @llvm.dx.flattened.thread.id.in.group()
-//CHECK-NEXT:   call void @"?main@@YAXI@Z"(i32 %0)
-//CHECK-NEXT:   ret void

clang/test/CodeGenHLSL/GlobalDestructors.hlsl

-// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.0-compute -std=hlsl202x -S -emit-llvm -disable-llvm-passes %s -o - | FileCheck %s --check-prefixes=CS,CHECK
-// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.3-library -std=hlsl202x -S -emit-llvm -disable-llvm-passes %s -o - | FileCheck %s --check-prefixes=LIB,CHECK
-
-// Make sure global variable for dtors exist for lib profile.
-// LIB:@llvm.global_dtors
-// Make sure global variable for dtors removed for compute profile.
-// CS-NOT:llvm.global_dtors
-
-struct Tail {
-  Tail() {
-    add(1);
-  }
-
-  ~Tail() {
-    add(-1);
-  }
-
-  void add(int V) {
-    static int Count = 0;
-    Count += V;
-  }
-};
-
-struct Pupper {
-  static int Count;
-
-  Pupper() {
-    Count += 1; // :)
-  }
-
-  ~Pupper() {
-    Count -= 1; // :(
-  }
-} GlobalPup;
-
-void Wag() {
-  static Tail T;
-  T.add(0);
-}
-
-int Pupper::Count = 0;
-
-[numthreads(1,1,1)]
-void main(unsigned GI : SV_GroupIndex) {
-  Wag();
-}
-
-// Make sure global variable for ctors/dtors removed.
-// CHECK-NOT:@llvm.global_ctors
-// CHECK-NOT:@llvm.global_dtors
-//CHECK:      define void @main()
-//CHECK-NEXT: entry:
-//CHECK-NEXT:   call void @_GLOBAL__sub_I_GlobalDestructors.hlsl()
-//CHECK-NEXT:   %0 = call i32 @llvm.dx.flattened.thread.id.in.group()
-//CHECK-NEXT:   call void @"?main@@YAXI@Z"(i32 %0)
-//CHECK-NEXT:   call void @_GLOBAL__D_a()
-//CHECK-NEXT:   ret void
-
-// This is really just a sanity check I needed for myself to verify that
-// function scope static variables also get destroyed properly.
-
-//CHECK: define internal void @_GLOBAL__D_a()
-//CHECK-NEXT: entry:
-//CHECK-NEXT:   call void @"??1Tail@@QAA@XZ"(ptr @"?T@?1??Wag@@YAXXZ@4UTail@@A")
-//CHECK-NEXT:   call void @"??1Pupper@@QAA@XZ"(ptr @"?GlobalPup@@3UPupper@@A")
-//CHECK-NEXT:   ret void

llvm/lib/Target/DirectX/CMakeLists.txt

add_llvm_target(DirectXCodeGen
   DirectXTargetMachine.cpp
   DXILMetadata.cpp
   DXILOpBuilder.cpp
   DXILOpLowering.cpp
   DXILPrepare.cpp
   DXILResource.cpp
   DXILResourceAnalysis.cpp
   DXILTranslateMetadata.cpp
+  GlobalCtorDtorCalls.cpp
   PointerTypeAnalysis.cpp
 
   LINK_COMPONENTS
   AsmPrinter
   Core
   SelectionDAG
   Support
   DirectXInfo

llvm/lib/Target/DirectX/DirectXTargetMachine.cpp

 //===----------------------------------------------------------------------===//
 
 #include "DirectXTargetMachine.h"
 #include "DXILResourceAnalysis.h"
 #include "DXILWriter/DXILWriterPass.h"
 #include "DirectX.h"
 #include "DirectXSubtarget.h"
 #include "DirectXTargetTransformInfo.h"
+#include "GlobalCtorDtorCalls.h"
 #include "TargetInfo/DirectXTargetInfo.h"
 #include "llvm/CodeGen/MachineModuleInfo.h"
 #include "llvm/CodeGen/Passes.h"
 #include "llvm/CodeGen/TargetPassConfig.h"
 #include "llvm/IR/IRPrintingPasses.h"
 #include "llvm/IR/LegacyPassManager.h"
 #include "llvm/MC/MCSectionDXContainer.h"
 #include "llvm/MC/SectionKind.h"

 void DirectXTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
   PB.registerPipelineParsingCallback(
       [](StringRef PassName, ModulePassManager &PM,
          ArrayRef<PassBuilder::PipelineElement>) {
         if (PassName == "print-dxil-resource") {
           PM.addPass(DXILResourcePrinterPass(dbgs()));
           return true;
         }
+        if (PassName == "global-ctor-dtor-calls") {
+          PM.addPass(GlobalCtorDtorCalls());
+          return true;
+        }
         return false;
       });
 
   PB.registerAnalysisRegistrationCallback([](ModuleAnalysisManager &MAM) {
     MAM.registerPass([&] { return DXILResourceAnalysis(); });
   });
+
+  PB.registerPipelineEarlySimplificationEPCallback(
+      [](ModulePassManager &PM, OptimizationLevel Level) {
+        PM.addPass(GlobalCtorDtorCalls());
+      });
 }
 
 bool DirectXTargetMachine::addPassesToEmitFile(
     PassManagerBase &PM, raw_pwrite_stream &Out, raw_pwrite_stream *DwoOut,
     CodeGenFileType FileType, bool DisableVerify,
     MachineModuleInfoWrapperPass *MMIWP) {
   TargetPassConfig *PassConfig = createPassConfig(PM);
   PassConfig->addCodeGenPrepare();

llvm/lib/Target/DirectX/GlobalCtorDtorCalls.h

+//===- Target/DirectX/GlobalCtorDtorCalls.h - calls on global ctor/dtor ---===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Transform pass to generate calls on global ctor and dtor.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_TARGET_DIRECTX__GLOBALCTORDTORCALLS_H
+#define LLVM_TARGET_DIRECTX__GLOBALCTORDTORCALLS_H
+
+#include "llvm/IR/PassManager.h"
+
+namespace llvm {
+
+/// Transform pass to generate calls on global ctors and dtors.
+class GlobalCtorDtorCalls : public PassInfoMixin<GlobalCtorDtorCalls> {
+
+public:
+  explicit GlobalCtorDtorCalls() {}
+  PreservedAnalyses run(Module &M, ModuleAnalysisManager &AM);
+};
+
+} // namespace llvm
+
+#endif

llvm/lib/Target/DirectX/GlobalCtorDtorCalls.cpp

+//===- Target/DirectX/GlobalCtorDtorCalls.cpp - calls on global ctor/dtor -===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// Transform pass to generate calls on global ctor and dtor.
+//
+//===----------------------------------------------------------------------===//
+
+#include "GlobalCtorDtorCalls.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/Triple.h"
+#include "llvm/IR/BasicBlock.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/IRBuilder.h"
+
+using namespace llvm;
+
+static void gatherFunctions(SmallVectorImpl<Function *> &Fns, llvm::Module &M,
+                            bool CtorOrDtor) {
+  const auto *GV =
+      M.getNamedGlobal(CtorOrDtor ? "llvm.global_ctors" : "llvm.global_dtors");
+  if (!GV)
+    return;
+  const auto *CA = dyn_cast<ConstantArray>(GV->getInitializer());
+  if (!CA)
+    return;
+
+  // The global_ctor array elements are a struct [Priority, Fn *, COMDat].
+  // HLSL neither supports priorities or COMDat values, so we will check those
+  // in an assert but not handle them.
+
+  llvm::SmallVector<Function *> CtorFns;
+  for (const auto &Ctor : CA->operands()) {
+    if (isa<ConstantAggregateZero>(Ctor))
+      continue;
+    ConstantStruct *CS = cast<ConstantStruct>(Ctor);
+
+    assert(cast<ConstantInt>(CS->getOperand(0))->getValue() == 65535 &&
+           "HLSL doesn't support setting priority for global ctors.");
+    assert(isa<ConstantPointerNull>(CS->getOperand(2)) &&
+           "HLSL doesn't support COMDat for global ctors.");
+    Fns.push_back(cast<Function>(CS->getOperand(1)));
+  }
+}
+
+PreservedAnalyses GlobalCtorDtorCalls::run(Module &M,
+                                           ModuleAnalysisManager &AM) {
+  SmallVector<Function *> CtorFns;
+  SmallVector<Function *> DtorFns;
+  gatherFunctions(CtorFns, M, true);
+  gatherFunctions(DtorFns, M, false);
+
+  // Insert a call to the global constructor at the beginning of the entry block
+  // to externally exported functions. This is a bit of a hack, but HLSL allows
+  // global constructors, but doesn't support driver initialization of globals.
+  for (auto &F : M.functions()) {
+    if (!F.hasFnAttribute("hlsl.shader"))
+      continue;
+    IRBuilder<> B(&F.getEntryBlock(), F.getEntryBlock().begin());
+    for (auto *Fn : CtorFns)
+      B.CreateCall(FunctionCallee(Fn));
+
+    // Insert global dtors before the terminator of the last instruction
+    B.SetInsertPoint(F.back().getTerminator());
+    for (auto *Fn : DtorFns)
+      B.CreateCall(FunctionCallee(Fn));
+  }
+
+  // No need to keep global ctors/dtors for non-lib profile after call to
+  // ctors/dtors added for entry.
+  Triple T(M.getTargetTriple());
+  if (T.getEnvironment() != Triple::EnvironmentType::Library) {
+    if (auto *GV = M.getNamedGlobal("llvm.global_ctors"))
+      GV->eraseFromParent();
+    if (auto *GV = M.getNamedGlobal("llvm.global_dtors"))
+      GV->eraseFromParent();
+  }
+
+  return PreservedAnalyses::all();
+}

llvm/test/CodeGen/DirectX/global_constructor_cs.ll

+; RUN: opt -S -passes="global-ctor-dtor-calls" < %s | FileCheck %s
+
+target datalayout = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64"
+target triple = "dxil-unknown-shadermodel6.0-compute"
+
+; Make sure global variable for ctors/dtors removed.
+; CHECK-NOT:@llvm.global_ctors
+; CHECK-NOT:@llvm.global_dtors
+
+
+%"class.hlsl::RWBuffer" = type { ptr }
+
+@"?Buffer@@3V?$RWBuffer@M@hlsl@@A" = global %"class.hlsl::RWBuffer" zeroinitializer, align 4
[e-mail]_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @_GLOBAL__sub_I_GlobalConstructors.hlsl, ptr null }]
+
+; Function Attrs: nounwind
+define internal void @"??__EBuffer@@YAXXZ"() #0 {
+entry:
+  %call = call noundef ptr @"??0?$RWBuffer@M@hlsl@@QAA@XZ"(ptr noundef nonnull align 4 dereferenceable(4) @"?Buffer@@3V?$RWBuffer@M@hlsl@@A")
+  ret void
+}
+
+; Function Attrs: inlinehint nounwind
+define linkonce_odr noundef ptr @"??0?$RWBuffer@M@hlsl@@QAA@XZ"(ptr noundef nonnull returned align 4 dereferenceable(4) %this) unnamed_addr #1 align 2 {
+entry:
+  %this.addr = alloca ptr, align 4
+  store ptr %this, ptr %this.addr, align 4
+  %this1 = load ptr, ptr %this.addr, align 4
+  %0 = call ptr @llvm.dx.create.handle(i8 1)
+  %h = getelementptr inbounds %"class.hlsl::RWBuffer", ptr %this1, i32 0, i32 0
+  store ptr %0, ptr %h, align 4
+  ret ptr %this1
+}
+
+; Function Attrs: norecurse nounwind
+define internal void @"?main@@YAXI@Z"(i32 noundef %GI) #2 {
+entry:
+  %GI.addr = alloca i32, align 4
+  store i32 %GI, ptr %GI.addr, align 4
+  ret void
+}
+
+;CHECK:      define void @main()
+;CHECK-NEXT: entry:
+;CHECK-NEXT:   call void @_GLOBAL__sub_I_GlobalConstructors.hlsl()
+;CHECK-NEXT:   %0 = call i32 @llvm.dx.flattened.thread.id.in.group()
+;CHECK-NEXT:   call void @"?main@@YAXI@Z"(i32 %0)
+;CHECK-NEXT:   ret void
+; Function Attrs: norecurse
+define void @main() #3 {
+entry:
+  %0 = call i32 @llvm.dx.flattened.thread.id.in.group()
+  call void @"?main@@YAXI@Z"(i32 %0)
+  ret void
+}
+
+; Function Attrs: nounwind readnone willreturn
+declare i32 @llvm.dx.flattened.thread.id.in.group() #4
+
+; Function Attrs: nounwind willreturn
+declare ptr @llvm.dx.create.handle(i8) #5
+
+; Function Attrs: nounwind
+define internal void @_GLOBAL__sub_I_GlobalConstructors.hlsl() #0 {
+entry:
+  call void @"??__EBuffer@@YAXXZ"()
+  ret void
+}
+
+attributes #0 = { nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #1 = { inlinehint nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #2 = { norecurse nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #3 = { norecurse "frame-pointer"="all" "hlsl.numthreads"="1,1,1" "hlsl.shader"="compute" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #4 = { nounwind readnone willreturn }
+attributes #5 = { nounwind willreturn }
+
+!hlsl.uavs = !{!0}
+!llvm.module.flags = !{!1, !2}
+!dx.valver = !{!3}
+
+
+!0 = !{ptr @"?Buffer@@3V?$RWBuffer@M@hlsl@@A", !"RWBuffer<float>", i32 0}
+!1 = !{i32 1, !"wchar_size", i32 4}
+!2 = !{i32 7, !"frame-pointer", i32 2}
+!3 = !{i32 1, i32 7}

llvm/test/CodeGen/DirectX/global_constructor_function.ll

+; RUN: opt -S -passes="global-ctor-dtor-calls" < %s | FileCheck %s
+
+target datalayout = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64"
+target triple = "dxil-unknown-shadermodel6.0-compute"
+
+; Make sure global variable for ctors/dtors removed.
+; CHECK-NOT:@llvm.global_ctors
+; CHECK-NOT:@llvm.global_dtors
+
+@"?i@@3HA" = global i32 0, align 4
[e-mail]_ctors = appending global [2 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @"?call_me_first@@YAXXZ", ptr null }, { i32, ptr, ptr } { i32 65535, ptr @"?then_call_me@@YAXXZ", ptr null }]
[e-mail]_dtors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @"?call_me_last@@YAXXZ", ptr null }]
+
+; Function Attrs: nounwind
+define void @"?call_me_first@@YAXXZ"() #0 {
+entry:
+  store i32 12, ptr @"?i@@3HA", align 4
+  ret void
+}
+
+; Function Attrs: nounwind
+define void @"?then_call_me@@YAXXZ"() #0 {
+entry:
+  store i32 12, ptr @"?i@@3HA", align 4
+  ret void
+}
+
+; Function Attrs: nounwind
+define void @"?call_me_last@@YAXXZ"() #0 {
+entry:
+  store i32 0, ptr @"?i@@3HA", align 4
+  ret void
+}
+
+; Function Attrs: norecurse nounwind
+define internal void @"?main@@YAXI@Z"(i32 noundef %GI) #1 {
+entry:
+  %GI.addr = alloca i32, align 4
+  store i32 %GI, ptr %GI.addr, align 4
+  ret void
+}
+;CHECK: define void @main()
+;CHECK-NEXT: entry:
+;CHECK-NEXT:   call void @"?call_me_first@@YAXXZ"()
+;CHECK-NEXT:   call void @"?then_call_me@@YAXXZ"()
+;CHECK-NEXT:   %0 = call i32 @llvm.dx.flattened.thread.id.in.group()
+;CHECK-NEXT:   call void @"?main@@YAXI@Z"(i32 %0)
+;CHECK-NEXT:   call void @"?call_me_last@@YAXXZ"(
+;CHECK-NEXT:   ret void
+; Function Attrs: norecurse
+define void @main() #2 {
+entry:
+  %0 = call i32 @llvm.dx.flattened.thread.id.in.group()
+  call void @"?main@@YAXI@Z"(i32 %0)
+  ret void
+}
+
+; Function Attrs: nounwind readnone willreturn
+declare i32 @llvm.dx.flattened.thread.id.in.group() #3
+
+attributes #0 = { nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #1 = { norecurse nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #2 = { norecurse "frame-pointer"="all" "hlsl.numthreads"="1,1,1" "hlsl.shader"="compute" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #3 = { nounwind readnone willreturn }
+
+!llvm.module.flags = !{!0, !1}
+!dx.valver = !{!2}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 7, !"frame-pointer", i32 2}
+!2 = !{i32 1, i32 7}

llvm/test/CodeGen/DirectX/global_constructor_lib.ll

+; RUN: opt -S -passes="global-ctor-dtor-calls" < %s | FileCheck %s
+
+target datalayout = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64"
+target triple = "dxil-unknown-shadermodel6.3-library"
+
+; Make sure global variable for ctors exist for lib profile.
+; CHECK:@llvm.global_ctors
+
+%"class.hlsl::RWBuffer" = type { ptr }
+
+@"?Buffer@@3V?$RWBuffer@M@hlsl@@A" = global %"class.hlsl::RWBuffer" zeroinitializer, align 4
[e-mail]_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @_GLOBAL__sub_I_GlobalConstructorLib.hlsl, ptr null }]
+
+; Function Attrs: nounwind
+define internal void @"??__EBuffer@@YAXXZ"() #0 {
+entry:
+  %call = call noundef ptr @"??0?$RWBuffer@M@hlsl@@QAA@XZ"(ptr noundef nonnull align 4 dereferenceable(4) @"?Buffer@@3V?$RWBuffer@M@hlsl@@A")
+  ret void
+}
+
+; Function Attrs: inlinehint nounwind
+define linkonce_odr noundef ptr @"??0?$RWBuffer@M@hlsl@@QAA@XZ"(ptr noundef nonnull returned align 4 dereferenceable(4) %this) unnamed_addr #1 align 2 {
+entry:
+  %this.addr = alloca ptr, align 4
+  store ptr %this, ptr %this.addr, align 4
+  %this1 = load ptr, ptr %this.addr, align 4
+  %0 = call ptr @llvm.dx.create.handle(i8 1)
+  %h = getelementptr inbounds %"class.hlsl::RWBuffer", ptr %this1, i32 0, i32 0
+  store ptr %0, ptr %h, align 4
+  ret ptr %this1
+}
+
+; Function Attrs: nounwind
+define internal void @"?FirstEntry@@YAXXZ"() #0 {
+entry:
+  ret void
+}
+; CHECK: define void @FirstEntry()
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   call void @_GLOBAL__sub_I_GlobalConstructorLib.hlsl()
+define void @FirstEntry() #2 {
+entry:
+  call void @"?FirstEntry@@YAXXZ"()
+  ret void
+}
+
+; Function Attrs: nounwind
+define internal void @"?SecondEntry@@YAXXZ"() #0 {
+entry:
+  ret void
+}
+; CHECK: define void @SecondEntry()
+; CHECK-NEXT: entry:
+; CHECK-NEXT:   call void @_GLOBAL__sub_I_GlobalConstructorLib.hlsl()
+; CHECK-NEXT:   call void @"?SecondEntry@@YAXXZ"()
+define void @SecondEntry() #2 {
+entry:
+  call void @"?SecondEntry@@YAXXZ"()
+  ret void
+}
+
+; Function Attrs: nounwind willreturn
+declare ptr @llvm.dx.create.handle(i8) #3
+
+; Function Attrs: nounwind
+define internal void @_GLOBAL__sub_I_GlobalConstructorLib.hlsl() #0 {
+entry:
+  call void @"??__EBuffer@@YAXXZ"()
+  ret void
+}
+
+attributes #0 = { nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #1 = { inlinehint nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #2 = { "frame-pointer"="all" "hlsl.numthreads"="1,1,1" "hlsl.shader"="compute" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #3 = { nounwind willreturn }
+
+!hlsl.uavs = !{!0}
+!llvm.module.flags = !{!1, !2}
+!dx.valver = !{!3}
+
+!0 = !{ptr @"?Buffer@@3V?$RWBuffer@M@hlsl@@A", !"RWBuffer<float>", i32 0}
+!1 = !{i32 1, !"wchar_size", i32 4}
+!2 = !{i32 7, !"frame-pointer", i32 2}
+!3 = !{i32 1, i32 7}

llvm/test/CodeGen/DirectX/global_destructor_cs.ll

+; RUN: opt -S -passes="global-ctor-dtor-calls" < %s | FileCheck %s
+
+target datalayout = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64"
+target triple = "dxil-unknown-shadermodel6.0-compute"
+
+; Make sure global variable for ctors/dtors removed.
+; CHECK-NOT:@llvm.global_ctors
+; CHECK-NOT:@llvm.global_dtors
+
+;CHECK:      define void @main()
+;CHECK-NEXT: entry:
+;CHECK-NEXT:   call void @_GLOBAL__sub_I_GlobalDestructors.hlsl()
+;CHECK-NEXT:   %0 = call i32 @llvm.dx.flattened.thread.id.in.group()
+;CHECK-NEXT:   call void @"?main@@YAXI@Z"(i32 %0)
+;CHECK-NEXT:   call void @_GLOBAL__D_a()
+;CHECK-NEXT:   ret void
+
+; This is really just a sanity check I needed for myself to verify that
+; function scope static variables also get destroyed properly.
+
+;CHECK: define internal void @_GLOBAL__D_a()
+;CHECK-NEXT: entry:
+;CHECK-NEXT:   call void @"??1Tail@@QAA@XZ"(ptr @"?T@?1??Wag@@YAXXZ@4UTail@@A")
+;CHECK-NEXT:   call void @"??1Pupper@@QAA@XZ"(ptr @"?GlobalPup@@3UPupper@@A")
+;CHECK-NEXT:   ret void
+
+%struct.Pupper = type { i8 }
+%struct.Tail = type { i8 }
+
+$"??1Pupper@@QAA@XZ" = comdat any
+
+$"??1Tail@@QAA@XZ" = comdat any
+
+@"?GlobalPup@@3UPupper@@A" = global %struct.Pupper zeroinitializer, align 1
+@"?T@?1??Wag@@YAXXZ@4UTail@@A" = internal global %struct.Tail zeroinitializer, align 1
+@"?$TSS0@?1??Wag@@YAXXZ@4HA" = internal global i32 0, align 4
+@_Init_thread_epoch = external thread_local global i32, align 4
+@"?Count@Pupper@@2HA" = global i32 0, align 4
+@"?Count@?1??add@Tail@@QAAXH@Z@4HA" = linkonce_odr global i32 0, align 4
[e-mail]_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @_GLOBAL__sub_I_GlobalDestructors.hlsl, ptr null }]
[e-mail]_dtors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @_GLOBAL__D_a, ptr null }]
+
+; Function Attrs: nounwind
+define internal void @"??__EGlobalPup@@YAXXZ"() #0 {
+entry:
+  %call = call noundef ptr @"??0Pupper@@QAA@XZ"(ptr noundef nonnull align 1 dereferenceable(1) @"?GlobalPup@@3UPupper@@A")
+  ret void
+}
+
+; Function Attrs: nounwind
+define linkonce_odr noundef ptr @"??0Pupper@@QAA@XZ"(ptr noundef nonnull returned align 1 dereferenceable(1) %this) unnamed_addr #0 align 2 {
+entry:
+  %this.addr = alloca ptr, align 4
+  store ptr %this, ptr %this.addr, align 4
+  %this1 = load ptr, ptr %this.addr, align 4
+  %0 = load i32, ptr @"?Count@Pupper@@2HA", align 4
+  %add = add nsw i32 %0, 1
+  store i32 %add, ptr @"?Count@Pupper@@2HA", align 4
+  ret ptr %this1
+}
+
+; Function Attrs: nounwind
+define linkonce_odr void @"??1Pupper@@QAA@XZ"(ptr noundef nonnull align 1 dereferenceable(1) %this) unnamed_addr #0 comdat align 2 {
+entry:
+  %this.addr = alloca ptr, align 4
+  store ptr %this, ptr %this.addr, align 4
+  %this1 = load ptr, ptr %this.addr, align 4
+  %0 = load i32, ptr @"?Count@Pupper@@2HA", align 4
+  %sub = sub nsw i32 %0, 1
+  store i32 %sub, ptr @"?Count@Pupper@@2HA", align 4
+  ret void
+}
+
+; Function Attrs: nounwind
+define void @"?Wag@@YAXXZ"() #0 {
+entry:
+  %0 = load atomic i32, ptr @"?$TSS0@?1??Wag@@YAXXZ@4HA" unordered, align 4
+  %1 = load i32, ptr @_Init_thread_epoch, align 4
+  %2 = icmp sgt i32 %0, %1
+  br i1 %2, label %init.attempt, label %init.end
+
+init.attempt:                                     ; preds = %entry
+  call void @_Init_thread_header(ptr @"?$TSS0@?1??Wag@@YAXXZ@4HA") #1
+  %3 = load atomic i32, ptr @"?$TSS0@?1??Wag@@YAXXZ@4HA" unordered, align 4
+  %4 = icmp eq i32 %3, -1
+  br i1 %4, label %init, label %init.end
+
+init:                                             ; preds = %init.attempt
+  %call = call noundef ptr @"??0Tail@@QAA@XZ"(ptr noundef nonnull align 1 dereferenceable(1) @"?T@?1??Wag@@YAXXZ@4UTail@@A")
+  call void @_Init_thread_footer(ptr @"?$TSS0@?1??Wag@@YAXXZ@4HA") #1
+  br label %init.end
+
+init.end:                                         ; preds = %init, %init.attempt, %entry
+  call void @"?add@Tail@@QAAXH@Z"(ptr noundef nonnull align 1 dereferenceable(1) @"?T@?1??Wag@@YAXXZ@4UTail@@A", i32 noundef 0)
+  ret void
+}
+
+; Function Attrs: nounwind
+declare void @_Init_thread_header(ptr) #1
+
+; Function Attrs: nounwind
+define linkonce_odr noundef ptr @"??0Tail@@QAA@XZ"(ptr noundef nonnull returned align 1 dereferenceable(1) %this) unnamed_addr #0 align 2 {
+entry:
+  %this.addr = alloca ptr, align 4
+  store ptr %this, ptr %this.addr, align 4
+  %this1 = load ptr, ptr %this.addr, align 4
+  call void @"?add@Tail@@QAAXH@Z"(ptr noundef nonnull align 1 dereferenceable(1) %this1, i32 noundef 1)
+  ret ptr %this1
+}
+
+; Function Attrs: nounwind
+define linkonce_odr void @"??1Tail@@QAA@XZ"(ptr noundef nonnull align 1 dereferenceable(1) %this) unnamed_addr #0 comdat align 2 {
+entry:
+  %this.addr = alloca ptr, align 4
+  store ptr %this, ptr %this.addr, align 4
+  %this1 = load ptr, ptr %this.addr, align 4
+  call void @"?add@Tail@@QAAXH@Z"(ptr noundef nonnull align 1 dereferenceable(1) %this1, i32 noundef -1)
+  ret void
+}
+
+; Function Attrs: nounwind
+declare void @_Init_thread_footer(ptr) #1
+
+; Function Attrs: nounwind
+define linkonce_odr void @"?add@Tail@@QAAXH@Z"(ptr noundef nonnull align 1 dereferenceable(1) %this, i32 noundef %V) #0 align 2 {
+entry:
+  %V.addr = alloca i32, align 4
+  %this.addr = alloca ptr, align 4
+  store i32 %V, ptr %V.addr, align 4
+  store ptr %this, ptr %this.addr, align 4
+  %this1 = load ptr, ptr %this.addr, align 4
+  %0 = load i32, ptr %V.addr, align 4
+  %1 = load i32, ptr @"?Count@?1??add@Tail@@QAAXH@Z@4HA", align 4
+  %add = add nsw i32 %1, %0
+  store i32 %add, ptr @"?Count@?1??add@Tail@@QAAXH@Z@4HA", align 4
+  ret void
+}
+
+; Function Attrs: norecurse nounwind
+define internal void @"?main@@YAXI@Z"(i32 noundef %GI) #2 {
+entry:
+  %GI.addr = alloca i32, align 4
+  store i32 %GI, ptr %GI.addr, align 4
+  call void @"?Wag@@YAXXZ"()
+  ret void
+}
+
+; Function Attrs: norecurse
+define void @main() #3 {
+entry:
+  %0 = call i32 @llvm.dx.flattened.thread.id.in.group()
+  call void @"?main@@YAXI@Z"(i32 %0)
+  ret void
+}
+
+; Function Attrs: nounwind readnone willreturn
+declare i32 @llvm.dx.flattened.thread.id.in.group() #4
+
+; Function Attrs: nounwind
+define internal void @_GLOBAL__sub_I_GlobalDestructors.hlsl() #0 {
+entry:
+  call void @"??__EGlobalPup@@YAXXZ"()
+  ret void
+}
+
+; Function Attrs: nounwind
+define internal void @_GLOBAL__D_a() #0 {
+entry:
+  call void @"??1Tail@@QAA@XZ"(ptr @"?T@?1??Wag@@YAXXZ@4UTail@@A")
+  call void @"??1Pupper@@QAA@XZ"(ptr @"?GlobalPup@@3UPupper@@A")
+  ret void
+}
+
+attributes #0 = { nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #1 = { nounwind }
+attributes #2 = { norecurse nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #3 = { norecurse "frame-pointer"="all" "hlsl.numthreads"="1,1,1" "hlsl.shader"="compute" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #4 = { nounwind readnone willreturn }
+
+!llvm.module.flags = !{!0, !1}
+!dx.valver = !{!2}
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 7, !"frame-pointer", i32 2}
+!2 = !{i32 1, i32 7}

llvm/test/CodeGen/DirectX/global_destructor_lib.ll

+; RUN: opt -S -passes="global-ctor-dtor-calls" < %s | FileCheck %s
+
+target datalayout = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64"
+target triple = "dxil-unknown-shadermodel6.3-library"
+
+; Make sure global variable for ctors/dtors exist.
+; CHECK:@llvm.global_ctors
+; CHECK:@llvm.global_dtors
+
+;CHECK:      define void @main()
+;CHECK-NEXT: entry:
+;CHECK-NEXT:   call void @_GLOBAL__sub_I_GlobalDestructors.hlsl()
+;CHECK-NEXT:   %0 = call i32 @llvm.dx.flattened.thread.id.in.group()
+;CHECK-NEXT:   call void @"?main@@YAXI@Z"(i32 %0)
+;CHECK-NEXT:   call void @_GLOBAL__D_a()
+;CHECK-NEXT:   ret void
+
+; This is really just a sanity check I needed for myself to verify that
+; function scope static variables also get destroyed properly.
+
+;CHECK: define internal void @_GLOBAL__D_a()
+;CHECK-NEXT: entry:
+;CHECK-NEXT:   call void @"??1Tail@@QAA@XZ"(ptr @"?T@?1??Wag@@YAXXZ@4UTail@@A")
+;CHECK-NEXT:   call void @"??1Pupper@@QAA@XZ"(ptr @"?GlobalPup@@3UPupper@@A")
+;CHECK-NEXT:   ret void
+
+%struct.Pupper = type { i8 }
+%struct.Tail = type { i8 }
+
+$"??1Pupper@@QAA@XZ" = comdat any
+
+$"??1Tail@@QAA@XZ" = comdat any
+
+@"?GlobalPup@@3UPupper@@A" = global %struct.Pupper zeroinitializer, align 1
+@"?T@?1??Wag@@YAXXZ@4UTail@@A" = internal global %struct.Tail zeroinitializer, align 1
+@"?$TSS0@?1??Wag@@YAXXZ@4HA" = internal global i32 0, align 4
+@_Init_thread_epoch = external thread_local global i32, align 4
+@"?Count@Pupper@@2HA" = global i32 0, align 4
+@"?Count@?1??add@Tail@@QAAXH@Z@4HA" = linkonce_odr global i32 0, align 4
[e-mail]_ctors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @_GLOBAL__sub_I_GlobalDestructors.hlsl, ptr null }]
[e-mail]_dtors = appending global [1 x { i32, ptr, ptr }] [{ i32, ptr, ptr } { i32 65535, ptr @_GLOBAL__D_a, ptr null }]
+
+; Function Attrs: nounwind
+define internal void @"??__EGlobalPup@@YAXXZ"() #0 {
+entry:
+  %call = call noundef ptr @"??0Pupper@@QAA@XZ"(ptr noundef nonnull align 1 dereferenceable(1) @"?GlobalPup@@3UPupper@@A")
+  ret void
+}
+
+; Function Attrs: nounwind
+define linkonce_odr noundef ptr @"??0Pupper@@QAA@XZ"(ptr noundef nonnull returned align 1 dereferenceable(1) %this) unnamed_addr #0 align 2 {
+entry:
+  %this.addr = alloca ptr, align 4
+  store ptr %this, ptr %this.addr, align 4
+  %this1 = load ptr, ptr %this.addr, align 4
+  %0 = load i32, ptr @"?Count@Pupper@@2HA", align 4
+  %add = add nsw i32 %0, 1
+  store i32 %add, ptr @"?Count@Pupper@@2HA", align 4
+  ret ptr %this1
+}
+
+; Function Attrs: nounwind
+define linkonce_odr void @"??1Pupper@@QAA@XZ"(ptr noundef nonnull align 1 dereferenceable(1) %this) unnamed_addr #0 comdat align 2 {
+entry:
+  %this.addr = alloca ptr, align 4
+  store ptr %this, ptr %this.addr, align 4
+  %this1 = load ptr, ptr %this.addr, align 4
+  %0 = load i32, ptr @"?Count@Pupper@@2HA", align 4
+  %sub = sub nsw i32 %0, 1
+  store i32 %sub, ptr @"?Count@Pupper@@2HA", align 4
+  ret void
+}
+
+; Function Attrs: nounwind
+define void @"?Wag@@YAXXZ"() #0 {
+entry:
+  %0 = load atomic i32, ptr @"?$TSS0@?1??Wag@@YAXXZ@4HA" unordered, align 4
+  %1 = load i32, ptr @_Init_thread_epoch, align 4
+  %2 = icmp sgt i32 %0, %1
+  br i1 %2, label %init.attempt, label %init.end
+
+init.attempt:                                     ; preds = %entry
+  call void @_Init_thread_header(ptr @"?$TSS0@?1??Wag@@YAXXZ@4HA") #1
+  %3 = load atomic i32, ptr @"?$TSS0@?1??Wag@@YAXXZ@4HA" unordered, align 4
+  %4 = icmp eq i32 %3, -1
+  br i1 %4, label %init, label %init.end
+
+init:                                             ; preds = %init.attempt
+  %call = call noundef ptr @"??0Tail@@QAA@XZ"(ptr noundef nonnull align 1 dereferenceable(1) @"?T@?1??Wag@@YAXXZ@4UTail@@A")
+  call void @_Init_thread_footer(ptr @"?$TSS0@?1??Wag@@YAXXZ@4HA") #1
+  br label %init.end
+
+init.end:                                         ; preds = %init, %init.attempt, %entry
+  call void @"?add@Tail@@QAAXH@Z"(ptr noundef nonnull align 1 dereferenceable(1) @"?T@?1??Wag@@YAXXZ@4UTail@@A", i32 noundef 0)
+  ret void
+}
+
+; Function Attrs: nounwind
+declare void @_Init_thread_header(ptr) #1
+
+; Function Attrs: nounwind
+define linkonce_odr noundef ptr @"??0Tail@@QAA@XZ"(ptr noundef nonnull returned align 1 dereferenceable(1) %this) unnamed_addr #0 align 2 {
+entry:
+  %this.addr = alloca ptr, align 4
+  store ptr %this, ptr %this.addr, align 4
+  %this1 = load ptr, ptr %this.addr, align 4
+  call void @"?add@Tail@@QAAXH@Z"(ptr noundef nonnull align 1 dereferenceable(1) %this1, i32 noundef 1)
+  ret ptr %this1
+}
+
+; Function Attrs: nounwind
+define linkonce_odr void @"??1Tail@@QAA@XZ"(ptr noundef nonnull align 1 dereferenceable(1) %this) unnamed_addr #0 comdat align 2 {
+entry:
+  %this.addr = alloca ptr, align 4
+  store ptr %this, ptr %this.addr, align 4
+  %this1 = load ptr, ptr %this.addr, align 4
+  call void @"?add@Tail@@QAAXH@Z"(ptr noundef nonnull align 1 dereferenceable(1) %this1, i32 noundef -1)
+  ret void
+}
+
+; Function Attrs: nounwind
+declare void @_Init_thread_footer(ptr) #1
+
+; Function Attrs: nounwind
+define linkonce_odr void @"?add@Tail@@QAAXH@Z"(ptr noundef nonnull align 1 dereferenceable(1) %this, i32 noundef %V) #0 align 2 {
+entry:
+  %V.addr = alloca i32, align 4
+  %this.addr = alloca ptr, align 4
+  store i32 %V, ptr %V.addr, align 4
+  store ptr %this, ptr %this.addr, align 4
+  %this1 = load ptr, ptr %this.addr, align 4
+  %0 = load i32, ptr %V.addr, align 4
+  %1 = load i32, ptr @"?Count@?1??add@Tail@@QAAXH@Z@4HA", align 4
+  %add = add nsw i32 %1, %0
+  store i32 %add, ptr @"?Count@?1??add@Tail@@QAAXH@Z@4HA", align 4
+  ret void
+}
+
+; Function Attrs: norecurse nounwind
+define internal void @"?main@@YAXI@Z"(i32 noundef %GI) #2 {
+entry:
+  %GI.addr = alloca i32, align 4
+  store i32 %GI, ptr %GI.addr, align 4
+  call void @"?Wag@@YAXXZ"()
+  ret void
+}
+
+; Function Attrs: norecurse
+define void @main() #3 {
+entry:
+  %0 = call i32 @llvm.dx.flattened.thread.id.in.group()
+  call void @"?main@@YAXI@Z"(i32 %0)
+  ret void
+}
+
+; Function Attrs: nounwind readnone willreturn
+declare i32 @llvm.dx.flattened.thread.id.in.group() #4
+
+; Function Attrs: nounwind
+define internal void @_GLOBAL__sub_I_GlobalDestructors.hlsl() #0 {
+entry:
+  call void @"??__EGlobalPup@@YAXXZ"()
+  ret void
+}
+
+; Function Attrs: nounwind
+define internal void @_GLOBAL__D_a() #0 {
+entry:
+  call void @"??1Tail@@QAA@XZ"(ptr @"?T@?1??Wag@@YAXXZ@4UTail@@A")
+  call void @"??1Pupper@@QAA@XZ"(ptr @"?GlobalPup@@3UPupper@@A")
+  ret void
+}
+
+attributes #0 = { nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #1 = { nounwind }
+attributes #2 = { norecurse nounwind "frame-pointer"="all" "min-legal-vector-width"="0" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #3 = { norecurse "frame-pointer"="all" "hlsl.numthreads"="1,1,1" "hlsl.shader"="library" "no-trapping-math"="true" "stack-protector-buffer-size"="8" }
+attributes #4 = { nounwind readnone willreturn }
+
+!llvm.module.flags = !{!0, !1}
+!dx.valver = !{!2}
+
+
+!0 = !{i32 1, !"wchar_size", i32 4}
+!1 = !{i32 7, !"frame-pointer", i32 2}
+!2 = !{i32 1, i32 7}