[OpenMP][AMDGCN] Support OpenMP offloading for AMDGCN architecture - Part 3

[saiislam]

Summary

Provides AMDGCN and NVPTX specific specialization of getGPUWarpSize,
getGPUThreadID, and getGPUNumThreads methods. Adds tests for AMDGCN
codegen for these methods in generic and simd modes. Also changes the
precondition in InitTempAlloca to be slightly more permissive. Useful for
AMDGCN OpenMP codegen where allocas are created with a cast to an
address space.

[ABataev]

clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.h

25

final

clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.h

25

final

[saiislam]

Added final to specialized classes.

[saiislam]

Added InitTempAlloca address space cast.

[saiislam]

clang/lib/CodeGen/CGExpr.cpp

129

@arsenm , it wasn't possible to post D78495 and this patch separately due to their interdependence. Test "clang/test/OpenMP/amdgcn_target_init_temp_alloca.cpp" below checks for this change.

[saiislam]

Generalized regex in one of the test to pass harbormaster build.

[ABataev]

clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.cpp

52

Move these comments to the header instead and remove them from the .cpp module.

clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp

43

Same, move these comments to the header.

clang/test/OpenMP/amdgcn_target_codegen.cpp

14

entry: not always exist in the LLVM IR functions, better not to check for it.

20–21

Do not rely on names for the locals, some release builds do not generate them. Use regexps instead.

clang/test/OpenMP/amdgcn_target_init_temp_alloca.cpp

16–17

same, use regexps instead of %-like names

[saiislam]

Moved comments to header files. Used regexps for test cases.

[ABataev]

clang/lib/CodeGen/CGOpenMPRuntimeGPU.h

201–206

Add these notes to the specialized functions too. It is required for better doxygen docs

clang/test/OpenMP/amdgcn_target_codegen.cpp

18–20

Same, the order of these checks should be strict

35–36

It should not be CHECK-DAG, the order of these 2 instructions is defined and the second one should definitely follow the first one

clang/test/OpenMP/amdgcn_target_init_temp_alloca.cpp

12

entry: again

16–18

Do not rely on the names again. Even as part of regexps

18

Just CHECK

[saiislam]

Improved comment placement. Corrected check lines. Rebased.

[ABataev]

clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.h

30–31

These 2 lines can be removed, they do not add anything useful.

clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.h

30–31

These 2 lines can be removed, they do not add anything useful.

clang/test/OpenMP/amdgcn_target_codegen.cpp

20

Remove DAG

[saiislam]

Changes as requested.

[ABataev]

LG

clang/lib/CodeGen/CGExpr.cpp

 Address CodeGenFunction::CreateDefaultAlignTempAlloca(llvm::Type *Ty,
                                                       const Twine &Name) {
   CharUnits Align =
     CharUnits::fromQuantity(CGM.getDataLayout().getABITypeAlignment(Ty));
   return CreateTempAlloca(Ty, Align, Name);
 }
 
 void CodeGenFunction::InitTempAlloca(Address Var, llvm::Value *Init) {
-  assert(isa<llvm::AllocaInst>(Var.getPointer()));
-  auto *Store = new llvm::StoreInst(Init, Var.getPointer(), /*volatile*/ false,
+  auto *Alloca = Var.getPointer();
+  assert(isa<llvm::AllocaInst>(Alloca) ||
+         (isa<llvm::AddrSpaceCastInst>(Alloca) &&
+          isa<llvm::AllocaInst>(
+              cast<llvm::AddrSpaceCastInst>(Alloca)->getPointerOperand())));
+
+  auto *Store = new llvm::StoreInst(Init, Alloca, /*volatile*/ false,
                                     Var.getAlignment().getAsAlign());
   llvm::BasicBlock *Block = AllocaInsertPt->getParent();
   Block->getInstList().insertAfter(AllocaInsertPt->getIterator(), Store);
 }
 
 Address CodeGenFunction::CreateIRTemp(QualType Ty, const Twine &Name) {
   CharUnits Align = getContext().getTypeAlignInChars(Ty);
   return CreateTempAlloca(ConvertType(Ty), Align, Name);

clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.h

-//===----- CGOpenMPRuntimeNVPTX.h - Interface to OpenMP NVPTX Runtimes ----===//
+//===--- CGOpenMPRuntimeAMDGCN.h - Interface to OpenMP AMDGCN Runtimes ---===//
 //
 // 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
 //
 //===----------------------------------------------------------------------===//
 //
-// This provides a class for OpenMP runtime code generation specialized to NVPTX
-// targets from generalized CGOpenMPRuntimeGPU class.
+// This provides a class for OpenMP runtime code generation specialized to
+// AMDGCN targets from generalized CGOpenMPRuntimeGPU class.
 //
 //===----------------------------------------------------------------------===//
 
-#ifndef LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIMENVPTX_H
-#define LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIMENVPTX_H
+#ifndef LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIMEAMDGCN_H
+#define LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIMEAMDGCN_H
 
 #include "CGOpenMPRuntime.h"
 #include "CGOpenMPRuntimeGPU.h"
 #include "CodeGenFunction.h"
 #include "clang/AST/StmtOpenMP.h"
 
 namespace clang {
 namespace CodeGen {
 
-class CGOpenMPRuntimeNVPTX : public CGOpenMPRuntimeGPU {
+class CGOpenMPRuntimeAMDGCN final : public CGOpenMPRuntimeGPU {
 
 public:
-  explicit CGOpenMPRuntimeNVPTX(CodeGenModule &CGM);
+  explicit CGOpenMPRuntimeAMDGCN(CodeGenModule &CGM);
   llvm::Value *getGPUWarpSize(CodeGenFunction &CGF) override;
+  llvm::Value *getGPUThreadID(CodeGenFunction &CGF) override;
+  llvm::Value *getGPUNumThreads(CodeGenFunction &CGF) override;
 };
 
-} // CodeGen namespace.
-} // clang namespace.
+} // namespace CodeGen
+} // namespace clang
 
-#endif // LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIMENVPTX_H
+#endif // LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIMEAMDGCN_H

clang/lib/CodeGen/CGOpenMPRuntimeAMDGCN.cpp

+//===-- CGOpenMPRuntimeAMDGCN.cpp - Interface to OpenMP AMDGCN Runtimes --===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// This provides a class for OpenMP runtime code generation specialized to
+// AMDGCN targets from generalized CGOpenMPRuntimeGPU class.
+//
+//===----------------------------------------------------------------------===//
+
+#include "CGOpenMPRuntimeAMDGCN.h"
+#include "CGOpenMPRuntimeGPU.h"
+#include "CodeGenFunction.h"
+#include "clang/AST/Attr.h"
+#include "clang/AST/DeclOpenMP.h"
+#include "clang/AST/StmtOpenMP.h"
+#include "clang/AST/StmtVisitor.h"
+#include "clang/Basic/Cuda.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/IR/IntrinsicsAMDGPU.h"
+
+using namespace clang;
+using namespace CodeGen;
+using namespace llvm::omp;
+
+CGOpenMPRuntimeAMDGCN::CGOpenMPRuntimeAMDGCN(CodeGenModule &CGM)
+    : CGOpenMPRuntimeGPU(CGM) {
+  if (!CGM.getLangOpts().OpenMPIsDevice)
+    llvm_unreachable("OpenMP AMDGCN can only handle device code.");
+}
+
+/// Get the GPU warp size.
+llvm::Value *CGOpenMPRuntimeAMDGCN::getGPUWarpSize(CodeGenFunction &CGF) {
+  CGBuilderTy &Bld = CGF.Builder;
+  // return constant compile-time target-specific warp size
+  unsigned WarpSize = CGF.getTarget().getGridValue(llvm::omp::GV_Warp_Size);
+  return Bld.getInt32(WarpSize);
+}
+
+/// Get the id of the current thread on the GPU.
+llvm::Value *CGOpenMPRuntimeAMDGCN::getGPUThreadID(CodeGenFunction &CGF) {
+  CGBuilderTy &Bld = CGF.Builder;
+  llvm::Function *F =
+      CGF.CGM.getIntrinsic(llvm::Intrinsic::amdgcn_workitem_id_x);
+  return Bld.CreateCall(F, llvm::None, "nvptx_tid");
+}
+
+/// Get the maximum number of threads in a block of the GPU.
+llvm::Value *CGOpenMPRuntimeAMDGCN::getGPUNumThreads(CodeGenFunction &CGF) {
+  CGBuilderTy &Bld = CGF.Builder;
+  llvm::Module *M = &CGF.CGM.getModule();
+  const char *LocSize = "__ockl_get_local_size";
+  llvm::Function *F = M->getFunction(LocSize);
+  if (!F) {
+    F = llvm::Function::Create(
+        llvm::FunctionType::get(CGF.Int64Ty, {CGF.Int32Ty}, false),
+        llvm::GlobalVariable::ExternalLinkage, LocSize, &CGF.CGM.getModule());
+  }
+  return Bld.CreateTrunc(
+      Bld.CreateCall(F, {Bld.getInt32(0)}, "nvptx_num_threads"), CGF.Int32Ty);
+}

clang/lib/CodeGen/CGOpenMPRuntimeGPU.h

 //===------ CGOpenMPRuntimeGPU.h - Interface to OpenMP GPU Runtimes ------===//
 //
 // 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
 //
 //===----------------------------------------------------------------------===//
 //
 // This provides a generalized class for OpenMP runtime code generation
-// specialized by GPU target NVPTX.
+// specialized by GPU targets NVPTX and AMDGCN.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIMEGPU_H
 #define LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIMEGPU_H
 
 #include "CGOpenMPRuntime.h"
 #include "CodeGenFunction.h"
protected:
   /// For NVPTX target contains data about SPMD/Non-SPMD execution mode +
   /// Full/Lightweight runtime mode. Used for better optimization.
   unsigned getDefaultLocationReserved2Flags() const override;
 
 public:
   explicit CGOpenMPRuntimeGPU(CodeGenModule &CGM);
   void clear() override;
 
   /// Declare generalized virtual functions which need to be defined
-  /// by all specializations of OpenMPGPURuntime Targets.
+  /// by all specializations of OpenMPGPURuntime Targets like AMDGCN
+  /// and NVPTX.
   virtual llvm::Value *getGPUWarpSize(CodeGenFunction &CGF) = 0;
+  virtual llvm::Value *getGPUThreadID(CodeGenFunction &CGF) = 0;
+  virtual llvm::Value *getGPUNumThreads(CodeGenFunction &CGF) = 0;
 
   /// Emit call to void __kmpc_push_proc_bind(ident_t *loc, kmp_int32
   /// global_tid, int proc_bind) to generate code for 'proc_bind' clause.
   virtual void emitProcBindClause(CodeGenFunction &CGF,
                                   llvm::omp::ProcBindKind ProcBind,
                                   SourceLocation Loc) override;
 
   /// Emits call to void __kmpc_push_num_threads(ident_t *loc, kmp_int32

clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp

 //===---- CGOpenMPRuntimeGPU.cpp - Interface to OpenMP GPU Runtimes ----===//
 //
 // 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
 //
 //===----------------------------------------------------------------------===//
 //
 // This provides a generalized class for OpenMP runtime code generation
-// specialized by GPU target NVPTX.
+// specialized by GPU targets NVPTX and AMDGCN.
 //
 //===----------------------------------------------------------------------===//
 
 #include "CGOpenMPRuntimeGPU.h"
 #include "CGOpenMPRuntimeNVPTX.h"
 #include "CodeGenFunction.h"
 #include "clang/AST/Attr.h"
 #include "clang/AST/DeclOpenMP.h"
public:
   /// Returns the list of the escaped variables with the variably modified
   /// types.
   ArrayRef<const ValueDecl *> getEscapedVariableLengthDecls() const {
     return EscapedVariableLengthDecls.getArrayRef();
   }
 };
 } // anonymous namespace
 
-/// Get the id of the current thread on the GPU.
-static llvm::Value *getNVPTXThreadID(CodeGenFunction &CGF) {
-  return CGF.EmitRuntimeCall(
-      llvm::Intrinsic::getDeclaration(
-          &CGF.CGM.getModule(), llvm::Intrinsic::nvvm_read_ptx_sreg_tid_x),
-      "nvptx_tid");
-}
-
 /// Get the id of the warp in the block.
 /// We assume that the warp size is 32, which is always the case
 /// on the NVPTX device, to generate more efficient code.
 static llvm::Value *getNVPTXWarpID(CodeGenFunction &CGF) {
   CGBuilderTy &Bld = CGF.Builder;
   unsigned LaneIDBits =
       CGF.getTarget().getGridValue(llvm::omp::GV_Warp_Size_Log2);
-  return Bld.CreateAShr(getNVPTXThreadID(CGF), LaneIDBits, "nvptx_warp_id");
+  auto &RT = static_cast<CGOpenMPRuntimeGPU &>(CGF.CGM.getOpenMPRuntime());
+  return Bld.CreateAShr(RT.getGPUThreadID(CGF), LaneIDBits, "nvptx_warp_id");
 }
 
 /// Get the id of the current lane in the Warp.
 /// We assume that the warp size is 32, which is always the case
 /// on the NVPTX device, to generate more efficient code.
 static llvm::Value *getNVPTXLaneID(CodeGenFunction &CGF) {
   CGBuilderTy &Bld = CGF.Builder;
   unsigned LaneIDMask = CGF.getContext().getTargetInfo().getGridValue(
       llvm::omp::GV_Warp_Size_Log2_Mask);
-  return Bld.CreateAnd(getNVPTXThreadID(CGF), Bld.getInt32(LaneIDMask),
+  auto &RT = static_cast<CGOpenMPRuntimeGPU &>(CGF.CGM.getOpenMPRuntime());
+  return Bld.CreateAnd(RT.getGPUThreadID(CGF), Bld.getInt32(LaneIDMask),
                        "nvptx_lane_id");
 }
 
-/// Get the maximum number of threads in a block of the GPU.
-static llvm::Value *getNVPTXNumThreads(CodeGenFunction &CGF) {
-  return CGF.EmitRuntimeCall(
-      llvm::Intrinsic::getDeclaration(
-          &CGF.CGM.getModule(), llvm::Intrinsic::nvvm_read_ptx_sreg_ntid_x),
-      "nvptx_num_threads");
-}
-
 /// Get the value of the thread_limit clause in the teams directive.
 /// For the 'generic' execution mode, the runtime encodes thread_limit in
 /// the launch parameters, always starting thread_limit+warpSize threads per
 /// CTA. The threads in the last warp are reserved for master execution.
 /// For the 'spmd' execution mode, all threads in a CTA are part of the team.
 static llvm::Value *getThreadLimit(CodeGenFunction &CGF,
                                    bool IsInSPMDExecutionMode = false) {
   CGBuilderTy &Bld = CGF.Builder;
   auto &RT = static_cast<CGOpenMPRuntimeGPU &>(CGF.CGM.getOpenMPRuntime());
   return IsInSPMDExecutionMode
-             ? getNVPTXNumThreads(CGF)
-             : Bld.CreateNUWSub(getNVPTXNumThreads(CGF), RT.getGPUWarpSize(CGF),
-                                "thread_limit");
+             ? RT.getGPUNumThreads(CGF)
+             : Bld.CreateNUWSub(RT.getGPUNumThreads(CGF),
+                                RT.getGPUWarpSize(CGF), "thread_limit");
 }
 
 /// Get the thread id of the OMP master thread.
 /// The master thread id is the first thread (lane) of the last warp in the
 /// GPU block.  Warp size is assumed to be some power of 2.
 /// Thread id is 0 indexed.
 /// E.g: If NumThreads is 33, master id is 32.
 ///      If NumThreads is 64, master id is 32.
 ///      If NumThreads is 1024, master id is 992.
 static llvm::Value *getMasterThreadID(CodeGenFunction &CGF) {
   CGBuilderTy &Bld = CGF.Builder;
-  llvm::Value *NumThreads = getNVPTXNumThreads(CGF);
   auto &RT = static_cast<CGOpenMPRuntimeGPU &>(CGF.CGM.getOpenMPRuntime());
+  llvm::Value *NumThreads = RT.getGPUNumThreads(CGF);
   // We assume that the warp size is a power of 2.
   llvm::Value *Mask = Bld.CreateNUWSub(RT.getGPUWarpSize(CGF), Bld.getInt32(1));
 
   return Bld.CreateAnd(Bld.CreateNUWSub(NumThreads, Bld.getInt32(1)),
                        Bld.CreateNot(Mask), "master_tid");
 }
 
 CGOpenMPRuntimeGPU::WorkerFunctionState::WorkerFunctionState(
void CGOpenMPRuntimeGPU::emitNonSPMDEntryHeader(CodeGenFunction &CGF,
                                                   WorkerFunctionState &WST) {
   CGBuilderTy &Bld = CGF.Builder;
 
   llvm::BasicBlock *WorkerBB = CGF.createBasicBlock(".worker");
   llvm::BasicBlock *MasterCheckBB = CGF.createBasicBlock(".mastercheck");
   llvm::BasicBlock *MasterBB = CGF.createBasicBlock(".master");
   EST.ExitBB = CGF.createBasicBlock(".exit");
 
+  auto &RT = static_cast<CGOpenMPRuntimeGPU &>(CGF.CGM.getOpenMPRuntime());
   llvm::Value *IsWorker =
-      Bld.CreateICmpULT(getNVPTXThreadID(CGF), getThreadLimit(CGF));
+      Bld.CreateICmpULT(RT.getGPUThreadID(CGF), getThreadLimit(CGF));
   Bld.CreateCondBr(IsWorker, WorkerBB, MasterCheckBB);
 
   CGF.EmitBlock(WorkerBB);
   emitCall(CGF, WST.Loc, WST.WorkerFn);
   CGF.EmitBranch(EST.ExitBB);
 
   CGF.EmitBlock(MasterCheckBB);
   llvm::Value *IsMaster =
-      Bld.CreateICmpEQ(getNVPTXThreadID(CGF), getMasterThreadID(CGF));
+      Bld.CreateICmpEQ(RT.getGPUThreadID(CGF), getMasterThreadID(CGF));
   Bld.CreateCondBr(IsMaster, MasterBB, EST.ExitBB);
 
   CGF.EmitBlock(MasterBB);
   IsInTargetMasterThreadRegion = true;
   // SEQUENTIAL (MASTER) REGION START
   // First action in sequential region:
   // Initialize the state of the OpenMP runtime library on the GPU.
   // TODO: Optimize runtime initialization and pass in correct value.
void CGOpenMPRuntimeGPU::emitCriticalRegion(
     const RegionCodeGenTy &CriticalOpGen, SourceLocation Loc,
     const Expr *Hint) {
   llvm::BasicBlock *LoopBB = CGF.createBasicBlock("omp.critical.loop");
   llvm::BasicBlock *TestBB = CGF.createBasicBlock("omp.critical.test");
   llvm::BasicBlock *SyncBB = CGF.createBasicBlock("omp.critical.sync");
   llvm::BasicBlock *BodyBB = CGF.createBasicBlock("omp.critical.body");
   llvm::BasicBlock *ExitBB = CGF.createBasicBlock("omp.critical.exit");
 
+  auto &RT = static_cast<CGOpenMPRuntimeGPU &>(CGF.CGM.getOpenMPRuntime());
+
   // Get the mask of active threads in the warp.
   llvm::Value *Mask = CGF.EmitRuntimeCall(
       createNVPTXRuntimeFunction(OMPRTL_NVPTX__kmpc_warp_active_thread_mask));
   // Fetch team-local id of the thread.
-  llvm::Value *ThreadID = getNVPTXThreadID(CGF);
+  llvm::Value *ThreadID = RT.getGPUThreadID(CGF);
 
   // Get the width of the team.
-  llvm::Value *TeamWidth = getNVPTXNumThreads(CGF);
+  llvm::Value *TeamWidth = RT.getGPUNumThreads(CGF);
 
   // Initialize the counter variable for the loop.
   QualType Int32Ty =
       CGF.getContext().getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/0);
   Address Counter = CGF.CreateMemTemp(Int32Ty, "critical_counter");
   LValue CounterLVal = CGF.MakeAddrLValue(Counter, Int32Ty);
   CGF.EmitStoreOfScalar(llvm::Constant::getNullValue(CGM.Int32Ty), CounterLVal,
                         /*isInit=*/true);
if (!TransferMedium) {
     TransferMedium = new llvm::GlobalVariable(
         M, Ty, /*isConstant=*/false, llvm::GlobalVariable::CommonLinkage,
         llvm::Constant::getNullValue(Ty), TransferMediumName,
         /*InsertBefore=*/nullptr, llvm::GlobalVariable::NotThreadLocal,
         SharedAddressSpace);
     CGM.addCompilerUsedGlobal(TransferMedium);
   }
 
+  auto &RT = static_cast<CGOpenMPRuntimeGPU &>(CGF.CGM.getOpenMPRuntime());
   // Get the CUDA thread id of the current OpenMP thread on the GPU.
-  llvm::Value *ThreadID = getNVPTXThreadID(CGF);
+  llvm::Value *ThreadID = RT.getGPUThreadID(CGF);
   // nvptx_lane_id = nvptx_id % warpsize
   llvm::Value *LaneID = getNVPTXLaneID(CGF);
   // nvptx_warp_id = nvptx_id / warpsize
   llvm::Value *WarpID = getNVPTXWarpID(CGF);
 
   Address AddrReduceListArg = CGF.GetAddrOfLocalVar(&ReduceListArg);
   Address LocalReduceList(
       Bld.CreatePointerBitCastOrAddrSpaceCast(
void CGOpenMPRuntimeGPU::functionFinished(CodeGenFunction &CGF) {
   FunctionGlobalizedDecls.erase(CGF.CurFn);
   CGOpenMPRuntime::functionFinished(CGF);
 }
 
 void CGOpenMPRuntimeGPU::getDefaultDistScheduleAndChunk(
     CodeGenFunction &CGF, const OMPLoopDirective &S,
     OpenMPDistScheduleClauseKind &ScheduleKind,
     llvm::Value *&Chunk) const {
+  auto &RT = static_cast<CGOpenMPRuntimeGPU &>(CGF.CGM.getOpenMPRuntime());
   if (getExecutionMode() == CGOpenMPRuntimeGPU::EM_SPMD) {
     ScheduleKind = OMPC_DIST_SCHEDULE_static;
-    Chunk = CGF.EmitScalarConversion(getNVPTXNumThreads(CGF),
+    Chunk = CGF.EmitScalarConversion(
+        RT.getGPUNumThreads(CGF),
         CGF.getContext().getIntTypeForBitwidth(32, /*Signed=*/0),
         S.getIterationVariable()->getType(), S.getBeginLoc());
     return;
   }
   CGOpenMPRuntime::getDefaultDistScheduleAndChunk(
       CGF, S, ScheduleKind, Chunk);
 }
 

clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.h

 #include "CGOpenMPRuntime.h"
 #include "CGOpenMPRuntimeGPU.h"
 #include "CodeGenFunction.h"
 #include "clang/AST/StmtOpenMP.h"
 
 namespace clang {
 namespace CodeGen {
 
-class CGOpenMPRuntimeNVPTX : public CGOpenMPRuntimeGPU {
+class CGOpenMPRuntimeNVPTX final : public CGOpenMPRuntimeGPU {
 
 public:
   explicit CGOpenMPRuntimeNVPTX(CodeGenModule &CGM);
   llvm::Value *getGPUWarpSize(CodeGenFunction &CGF) override;
+  llvm::Value *getGPUThreadID(CodeGenFunction &CGF) override;
+  llvm::Value *getGPUNumThreads(CodeGenFunction &CGF) override;
 };
 
 } // CodeGen namespace.
 } // clang namespace.
 
 #endif // LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIMENVPTX_H

clang/lib/CodeGen/CGOpenMPRuntimeNVPTX.cpp

 
 /// Get the GPU warp size.
 llvm::Value *CGOpenMPRuntimeNVPTX::getGPUWarpSize(CodeGenFunction &CGF) {
   return CGF.EmitRuntimeCall(
       llvm::Intrinsic::getDeclaration(
           &CGF.CGM.getModule(), llvm::Intrinsic::nvvm_read_ptx_sreg_warpsize),
       "nvptx_warp_size");
 }
+
+/// Get the id of the current thread on the GPU.
+llvm::Value *CGOpenMPRuntimeNVPTX::getGPUThreadID(CodeGenFunction &CGF) {
+  CGBuilderTy &Bld = CGF.Builder;
+  llvm::Function *F;
+  F = llvm::Intrinsic::getDeclaration(
+      &CGF.CGM.getModule(), llvm::Intrinsic::nvvm_read_ptx_sreg_tid_x);
+  return Bld.CreateCall(F, llvm::None, "nvptx_tid");
+}
+
+/// Get the maximum number of threads in a block of the GPU.
+llvm::Value *CGOpenMPRuntimeNVPTX::getGPUNumThreads(CodeGenFunction &CGF) {
+  CGBuilderTy &Bld = CGF.Builder;
+  llvm::Function *F;
+  F = llvm::Intrinsic::getDeclaration(
+      &CGF.CGM.getModule(), llvm::Intrinsic::nvvm_read_ptx_sreg_ntid_x);
+  return Bld.CreateCall(F, llvm::None, "nvptx_num_threads");
+}

clang/lib/CodeGen/CMakeLists.txt

add_clang_library(clangCodeGen
   CGLoopInfo.cpp
   CGNonTrivialStruct.cpp
   CGObjC.cpp
   CGObjCGNU.cpp
   CGObjCMac.cpp
   CGObjCRuntime.cpp
   CGOpenCLRuntime.cpp
   CGOpenMPRuntime.cpp
+  CGOpenMPRuntimeAMDGCN.cpp
   CGOpenMPRuntimeGPU.cpp
   CGOpenMPRuntimeNVPTX.cpp
   CGRecordLayoutBuilder.cpp
   CGStmt.cpp
   CGStmtOpenMP.cpp
   CGVTT.cpp
   CGVTables.cpp
   CodeGenABITypes.cpp

clang/lib/CodeGen/CodeGenModule.cpp

 #include "CGBlocks.h"
 #include "CGCUDARuntime.h"
 #include "CGCXXABI.h"
 #include "CGCall.h"
 #include "CGDebugInfo.h"
 #include "CGObjCRuntime.h"
 #include "CGOpenCLRuntime.h"
 #include "CGOpenMPRuntime.h"
+#include "CGOpenMPRuntimeAMDGCN.h"
 #include "CGOpenMPRuntimeNVPTX.h"
 #include "CodeGenFunction.h"
 #include "CodeGenPGO.h"
 #include "ConstantEmitter.h"
 #include "CoverageMappingGen.h"
 #include "TargetInfo.h"
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/CharUnits.h"
void CodeGenModule::createOpenMPRuntime() {
   // If it does not exist use the default implementation.
   switch (getTriple().getArch()) {
   case llvm::Triple::nvptx:
   case llvm::Triple::nvptx64:
     assert(getLangOpts().OpenMPIsDevice &&
            "OpenMP NVPTX is only prepared to deal with device code.");
     OpenMPRuntime.reset(new CGOpenMPRuntimeNVPTX(*this));
     break;
+  case llvm::Triple::amdgcn:
+    assert(getLangOpts().OpenMPIsDevice &&
+           "OpenMP AMDGCN is only prepared to deal with device code.");
+    OpenMPRuntime.reset(new CGOpenMPRuntimeAMDGCN(*this));
+    break;
   default:
     if (LangOpts.OpenMPSimd)
       OpenMPRuntime.reset(new CGOpenMPSIMDRuntime(*this));
     else
       OpenMPRuntime.reset(new CGOpenMPRuntime(*this));
     break;
   }
 }

clang/test/OpenMP/amdgcn_target_codegen.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
+
+#define N 1000
+
+int test_amdgcn_target_tid_threads() {
+  // CHECK-LABEL: test_amdgcn_target_tid_threads
+  // CHECK-LABEL: entry:
+
+  int arr[N];
+
+// CHECK: %nvptx_num_threads = call i64 @__ockl_get_local_size(i32 0)
+// CHECK-DAG: [[VAR1:%[0-9]+]] = trunc i64 %nvptx_num_threads to i32
+// CHECK-DAG: %thread_limit = sub nuw i32 [[VAR1]], 64
+// CHECK-DAG: %nvptx_tid = call i32 @llvm.amdgcn.workitem.id.x()
+#pragma omp target
+  for (int i = 0; i < N; i++) {
+    arr[i] = 1;
+  }
+
+  return arr[0];
+}
+
+int test_amdgcn_target_tid_threads_simd() {
+  // CHECK-LABEL: test_amdgcn_target_tid_threads_simd
+  // CHECK-LABEL: entry:
+
+  int arr[N];
+
+// CHECK: %nvptx_num_threads = call i64 @__ockl_get_local_size(i32 0)
+// CHECK-DAG: [[VAR2:%[0-9]+]] = trunc i64 %nvptx_num_threads to i32
+// CHECK-DAG: call void @__kmpc_spmd_kernel_init(i32 [[VAR2]], i16 0, i16 0)
+#pragma omp target simd
+  for (int i = 0; i < N; i++) {
+    arr[i] = 1;
+  }
+  return arr[0];
+}
+
+#endif

clang/test/OpenMP/amdgcn_target_init_temp_alloca.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
+
+#define N 100
+
+int test_amdgcn_target_temp_alloca() {
+  // CHECK-LABEL: test_amdgcn_target_temp_alloca
+  // CHECK-LABEL: entry:
+
+  int arr[N];
+
+  // CHECK:      %arr.addr = alloca [100 x i32]*, align 8, addrspace(5)
+  // CHECK-NEXT: %arr.addr.ascast = addrspacecast [100 x i32]* addrspace(5)* %arr.addr to [100 x i32]**
+  // CHECK-DAG:  store [100 x i32]* %arr, [100 x i32]** %arr.addr.ascast, align 8
+
+#pragma omp target
+  for (int i = 0; i < N; i++) {
+    arr[i] = 1;
+  }
+
+  return arr[0];
+}