[OpenMP] Use the OpenMPIRBuilder for `omp parallel`

This allows to use the OpenMPIRBuilder for parallel regions. Code was
extracted from D61953 and adapted to work with the new version (D70109).

All but one feature should be supported. An update of this patch will
provide test coverage and privatization other than shared.

Reviewed By: fghanim

Differential Revision: https://reviews.llvm.org/D70290

1 files changed, 95 insertions, 0 deletions

diff --git a/clang/lib/CodeGen/CGStmtOpenMP.cpp b/clang/lib/CodeGen/CGStmtOpenMP.cpp
index a2a59b64ddb..a38a79b6454 100644
--- a/clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ b/clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -21,6 +21,7 @@
 #include "clang/AST/Stmt.h"
 #include "clang/AST/StmtOpenMP.h"
 #include "clang/Basic/PrettyStackTrace.h"
+#include "llvm/Frontend/OpenMP/OMPIRBuilder.h"
 using namespace clang;
 using namespace CodeGen;
 using namespace llvm::omp;
@@ -1318,6 +1319,87 @@ static void emitEmptyBoundParameters(CodeGenFunction &,
                                      llvm::SmallVectorImpl<llvm::Value *> &) {}
 
 void CodeGenFunction::EmitOMPParallelDirective(const OMPParallelDirective &S) {
+
+  if (llvm::OpenMPIRBuilder *OMPBuilder = CGM.getOpenMPIRBuilder()) {
+    // Check if we have any if clause associated with the directive.
+    llvm::Value *IfCond = nullptr;
+    if (const auto *C = S.getSingleClause<OMPIfClause>())
+      IfCond = EmitScalarExpr(C->getCondition(),
+                              /*IgnoreResultAssign=*/true);
+
+    llvm::Value *NumThreads = nullptr;
+    if (const auto *NumThreadsClause = S.getSingleClause<OMPNumThreadsClause>())
+      NumThreads = EmitScalarExpr(NumThreadsClause->getNumThreads(),
+                                  /*IgnoreResultAssign=*/true);
+
+    ProcBindKind ProcBind = OMP_PROC_BIND_default;
+    if (const auto *ProcBindClause = S.getSingleClause<OMPProcBindClause>())
+      ProcBind = ProcBindClause->getProcBindKind();
+
+    using InsertPointTy = llvm::OpenMPIRBuilder::InsertPointTy;
+
+    // The cleanup callback that finalizes all variabels at the given location,
+    // thus calls destructors etc.
+    auto FiniCB = [this](InsertPointTy IP) {
+      CGBuilderTy::InsertPointGuard IPG(Builder);
+      assert(IP.getBlock()->end() != IP.getPoint() &&
+             "OpenMP IR Builder should cause terminated block!");
+      llvm::BasicBlock *IPBB = IP.getBlock();
+      llvm::BasicBlock *DestBB = IPBB->splitBasicBlock(IP.getPoint());
+      IPBB->getTerminator()->eraseFromParent();
+      Builder.SetInsertPoint(IPBB);
+      CodeGenFunction::JumpDest Dest = getJumpDestInCurrentScope(DestBB);
+      EmitBranchThroughCleanup(Dest);
+    };
+
+    // Privatization callback that performs appropriate action for
+    // shared/private/firstprivate/lastprivate/copyin/... variables.
+    //
+    // TODO: This defaults to shared right now.
+    auto PrivCB = [](InsertPointTy AllocaIP, InsertPointTy CodeGenIP,
+                     llvm::Value &Val, llvm::Value *&ReplVal) {
+      // The next line is appropriate only for variables (Val) with the
+      // data-sharing attribute "shared".
+      ReplVal = &Val;
+
+      return CodeGenIP;
+    };
+
+    const CapturedStmt *CS = S.getCapturedStmt(OMPD_parallel);
+    const Stmt *ParallelRegionBodyStmt = CS->getCapturedStmt();
+
+    auto BodyGenCB = [ParallelRegionBodyStmt,
+                      this](InsertPointTy AllocaIP, InsertPointTy CodeGenIP,
+                            llvm::BasicBlock &ContinuationBB) {
+      auto OldAllocaIP = AllocaInsertPt;
+      AllocaInsertPt = &*AllocaIP.getPoint();
+
+      auto OldReturnBlock = ReturnBlock;
+      ReturnBlock = getJumpDestInCurrentScope(&ContinuationBB);
+
+      llvm::BasicBlock *CodeGenIPBB = CodeGenIP.getBlock();
+      CodeGenIPBB->splitBasicBlock(CodeGenIP.getPoint());
+      llvm::Instruction *CodeGenIPBBTI = CodeGenIPBB->getTerminator();
+      CodeGenIPBBTI->removeFromParent();
+
+      Builder.SetInsertPoint(CodeGenIPBB);
+
+      EmitStmt(ParallelRegionBodyStmt);
+
+      Builder.Insert(CodeGenIPBBTI);
+
+      AllocaInsertPt = OldAllocaIP;
+      ReturnBlock = OldReturnBlock;
+    };
+
+    CGCapturedStmtInfo CGSI(*CS, CR_OpenMP);
+    CodeGenFunction::CGCapturedStmtRAII CapInfoRAII(*this, &CGSI);
+    Builder.restoreIP(OMPBuilder->CreateParallel(Builder, BodyGenCB, PrivCB,
+                                                 FiniCB, IfCond, NumThreads,
+                                                 ProcBind, S.hasCancel()));
+    return;
+  }
+
   // Emit parallel region as a standalone region.
   auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &Action) {
     Action.Enter(CGF);
@@ -4747,6 +4829,19 @@ void CodeGenFunction::EmitOMPCancelDirective(const OMPCancelDirective &S) {
       break;
     }
   }
+  if (llvm::OpenMPIRBuilder *OMPBuilder = CGM.getOpenMPIRBuilder()) {
+    // TODO: This check is necessary as we only generate `omp parallel` through
+    // the OpenMPIRBuilder for now.
+    if (S.getCancelRegion() == OMPD_parallel) {
+      llvm::Value *IfCondition = nullptr;
+      if (IfCond)
+        IfCondition = EmitScalarExpr(IfCond,
+                                     /*IgnoreResultAssign=*/true);
+      return Builder.restoreIP(
+          OMPBuilder->CreateCancel(Builder, IfCondition, S.getCancelRegion()));
+    }
+  }
+
   CGM.getOpenMPRuntime().emitCancelCall(*this, S.getBeginLoc(), IfCond,
                                         S.getCancelRegion());
 }