[OpenMP][FIX] Allow device constructors for AMD GPU

[jdoerfert]

Summary

In AMD GPU device code the globals are in AS(1). Before, we crashed if
the global was a structure. Now we simply cast away the AS before we
generate the code to initialize the global.

[JonChesterfield]

Nice, thanks!

[ronlieb]

Hi Johannes,
assert in two of the tests

#7 0x00005576b8d055e4 llvm::ConstantExpr::getAddrSpaceCast(llvm::Constant*, llvm::Type*, bool) (/work/omp-vega20-0/openmp-offload-amdgpu-runtime/llvm.build/bin/clang-15+0x1e4b5e4)

https://lab.llvm.org/buildbot/#/builders/193/builds/8594/steps/6/logs/FAIL__libomptarget____x86_64-pc-linux-gnu__global_

[jdoerfert]

Fix AS(0) -> AS(0) cast error

[JonChesterfield]

Thanks!

[ronlieb]

i applied it locally and ran bot equiv, looks good to me

[thakis]

Looks like this breaks tests: http://45.33.8.238/linux/71358/step_7.txt

Please take a look and revert for now if it takes a while to fix.

[jdoerfert]

In D121837#3390042, @thakis wrote:

Looks like this breaks tests: http://45.33.8.238/linux/71358/step_7.txt

Please take a look and revert for now if it takes a while to fix.

I relaxed the check line. Should resolve the problem.

[kamaub]

The test cases added with this commit failed on clang-ppc64be-linux-lnt # 13809 could you please revert this change, and recommit with the test case corrected? thank you.

[jdoerfert]

In D121837#3392346, @kamaub wrote:

The test cases added with this commit failed on clang-ppc64be-linux-lnt # 13809 could you please revert this change, and recommit with the test case corrected? thank you.

[EDIT] I did in https://reviews.llvm.org/rGb4cc3b1dd8d7200640210513263b28187f810703, it seems the metadata is reordered in that bot... I am fixing it right now...

clang/lib/CodeGen/CGOpenMPRuntime.cpp

 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/SetOperations.h"
 #include "llvm/ADT/SmallBitVector.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/Bitcode/BitcodeReader.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/DerivedTypes.h"
 #include "llvm/IR/GlobalValue.h"
+#include "llvm/IR/InstrTypes.h"
 #include "llvm/IR/Value.h"
 #include "llvm/Support/AtomicOrdering.h"
 #include "llvm/Support/Format.h"
 #include "llvm/Support/raw_ostream.h"
 #include <cassert>
 #include <numeric>
 
 using namespace clang;
if (CGM.getLangOpts().OpenMPIsDevice) {
       const CGFunctionInfo &FI = CGM.getTypes().arrangeNullaryFunction();
       llvm::FunctionType *FTy = CGM.getTypes().GetFunctionType(FI);
       llvm::Function *Fn = CGM.CreateGlobalInitOrCleanUpFunction(
           FTy, Twine(Buffer, "_ctor"), FI, Loc);
       auto NL = ApplyDebugLocation::CreateEmpty(CtorCGF);
       CtorCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidTy, Fn, FI,
                             FunctionArgList(), Loc, Loc);
       auto AL = ApplyDebugLocation::CreateArtificial(CtorCGF);
+      auto *AddrInAS0 = llvm::ConstantExpr::getAddrSpaceCast(
+          Addr, llvm::PointerType::getWithSamePointeeType(
+                    cast<llvm::PointerType>(Addr->getType()), 0));
       CtorCGF.EmitAnyExprToMem(
-          Init, Address::deprecated(Addr, CGM.getContext().getDeclAlign(VD)),
+          Init,
+          Address::deprecated(AddrInAS0, CGM.getContext().getDeclAlign(VD)),
           Init->getType().getQualifiers(),
           /*IsInitializer=*/true);
       CtorCGF.FinishFunction();
       Ctor = Fn;
       ID = llvm::ConstantExpr::getBitCast(Fn, CGM.Int8PtrTy);
       CGM.addUsedGlobal(cast<llvm::GlobalValue>(Ctor));
     } else {
       Ctor = new llvm::GlobalVariable(
if (CGM.getLangOpts().OpenMPIsDevice) {
       llvm::Function *Fn = CGM.CreateGlobalInitOrCleanUpFunction(
           FTy, Twine(Buffer, "_dtor"), FI, Loc);
       auto NL = ApplyDebugLocation::CreateEmpty(DtorCGF);
       DtorCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidTy, Fn, FI,
                             FunctionArgList(), Loc, Loc);
       // Create a scope with an artificial location for the body of this
       // function.
       auto AL = ApplyDebugLocation::CreateArtificial(DtorCGF);
+      auto *AddrInAS0 = llvm::ConstantExpr::getAddrSpaceCast(
+          Addr, llvm::PointerType::getWithSamePointeeType(
+                    cast<llvm::PointerType>(Addr->getType()), 0));
       DtorCGF.emitDestroy(
-          Address::deprecated(Addr, CGM.getContext().getDeclAlign(VD)), ASTTy,
-          DtorCGF.getDestroyer(ASTTy.isDestructedType()),
+          Address::deprecated(AddrInAS0, CGM.getContext().getDeclAlign(VD)),
+          ASTTy, DtorCGF.getDestroyer(ASTTy.isDestructedType()),
           DtorCGF.needsEHCleanup(ASTTy.isDestructedType()));
       DtorCGF.FinishFunction();
       Dtor = Fn;
       ID = llvm::ConstantExpr::getBitCast(Fn, CGM.Int8PtrTy);
       CGM.addUsedGlobal(cast<llvm::GlobalValue>(Dtor));
     } else {
       Dtor = new llvm::GlobalVariable(
           CGM.getModule(), CGM.Int8Ty, /*isConstant=*/true,

clang/test/OpenMP/amdgcn_target_global_constructor.cpp

+// REQUIRES: amdgpu-registered-target
+
+// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm-bc %s -o %t-ppc-host.bc
+// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple amdgcn-amd-amdhsa -fopenmp-targets=amdgcn-amd-amdhsa -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s
+// expected-no-diagnostics
+#ifndef HEADER
+#define HEADER
+
+void foo(void);
+
+struct S {
+  int a;
+  S() : a(1) {}
+  ~S() { foo(); }
+};
+
+#pragma omp declare target
+S A;
+#pragma omp end declare target
+
+#endif