[Flang] Support generic execution of parallel regions

clang/lib/CodeGen/CGOpenMPRuntime.cpp

@@ -11354,8 +11354,8 @@ void CGOpenMPRuntime::emitTargetDataCalls(
   llvm::OpenMPIRBuilder::LocationDescription OmpLoc(CodeGenIP);
   llvm::OpenMPIRBuilder::InsertPointTy AfterIP =
       cantFail(OMPBuilder.createTargetData(
-          OmpLoc, AllocaIP, CodeGenIP, DeviceID, IfCondVal, Info, GenMapInfoCB,
-          CustomMapperCB,
+          OmpLoc, AllocaIP, CodeGenIP, /*DeallocIPs=*/{}, DeviceID, IfCondVal,
+          Info, GenMapInfoCB, CustomMapperCB,
           /*MapperFunc=*/nullptr, BodyCB, DeviceAddrCB, RTLoc));
   CGF.Builder.restoreIP(AfterIP);
 }

clang/lib/CodeGen/CGStmtOpenMP.cpp

@@ -2238,20 +2238,21 @@ void CodeGenFunction::EmitOMPParallelDirective(const OMPParallelDirective &S) {
     const CapturedStmt *CS = S.getCapturedStmt(OMPD_parallel);
     const Stmt *ParallelRegionBodyStmt = CS->getCapturedStmt();
 
-    auto BodyGenCB = [&, this](InsertPointTy AllocaIP,
-                               InsertPointTy CodeGenIP) {
+    auto BodyGenCB = [&, this](InsertPointTy AllocIP, InsertPointTy CodeGenIP,
+                               ArrayRef<InsertPointTy> DeallocIPs) {
       OMPBuilderCBHelpers::EmitOMPOutlinedRegionBody(
-          *this, ParallelRegionBodyStmt, AllocaIP, CodeGenIP, "parallel");
+          *this, ParallelRegionBodyStmt, AllocIP, CodeGenIP, "parallel");
       return llvm::Error::success();
     };
 
     CGCapturedStmtInfo CGSI(*CS, CR_OpenMP);
     CodeGenFunction::CGCapturedStmtRAII CapInfoRAII(*this, &CGSI);
     llvm::OpenMPIRBuilder::InsertPointTy AllocaIP(
         AllocaInsertPt->getParent(), AllocaInsertPt->getIterator());
-    llvm::OpenMPIRBuilder::InsertPointTy AfterIP = cantFail(
-        OMPBuilder.createParallel(Builder, AllocaIP, BodyGenCB, PrivCB, FiniCB,
-                                  IfCond, NumThreads, ProcBind, S.hasCancel()));
+    llvm::OpenMPIRBuilder::InsertPointTy AfterIP =
+        cantFail(OMPBuilder.createParallel(
+            Builder, AllocaIP, /*DeallocIPs=*/{}, BodyGenCB, PrivCB, FiniCB,
+            IfCond, NumThreads, ProcBind, S.hasCancel()));
     Builder.restoreIP(AfterIP);
     return;
   }
@@ -4936,21 +4937,23 @@ void CodeGenFunction::EmitOMPSectionsDirective(const OMPSectionsDirective &S) {
     llvm::SmallVector<BodyGenCallbackTy, 4> SectionCBVector;
     if (CS) {
       for (const Stmt *SubStmt : CS->children()) {
-        auto SectionCB = [this, SubStmt](InsertPointTy AllocaIP,
-                                         InsertPointTy CodeGenIP) {
-          OMPBuilderCBHelpers::EmitOMPInlinedRegionBody(
-              *this, SubStmt, AllocaIP, CodeGenIP, "section");
+        auto SectionCB = [this, SubStmt](InsertPointTy AllocIP,
+                                         InsertPointTy CodeGenIP,
+                                         ArrayRef<InsertPointTy> DeallocIPs) {
+          OMPBuilderCBHelpers::EmitOMPInlinedRegionBody(*this, SubStmt, AllocIP,
+                                                        CodeGenIP, "section");
           return llvm::Error::success();
         };
         SectionCBVector.push_back(SectionCB);
       }
     } else {
-      auto SectionCB = [this, CapturedStmt](InsertPointTy AllocaIP,
-                                            InsertPointTy CodeGenIP) {
-        OMPBuilderCBHelpers::EmitOMPInlinedRegionBody(
-            *this, CapturedStmt, AllocaIP, CodeGenIP, "section");
-        return llvm::Error::success();
-      };
+      auto SectionCB =
+          [this, CapturedStmt](InsertPointTy AllocIP, InsertPointTy CodeGenIP,
+                               ArrayRef<InsertPointTy> DeallocIPs) {
+            OMPBuilderCBHelpers::EmitOMPInlinedRegionBody(
+                *this, CapturedStmt, AllocIP, CodeGenIP, "section");
+            return llvm::Error::success();
+          };
       SectionCBVector.push_back(SectionCB);
     }
 
@@ -5004,10 +5007,11 @@ void CodeGenFunction::EmitOMPSectionDirective(const OMPSectionDirective &S) {
       return llvm::Error::success();
     };
 
-    auto BodyGenCB = [SectionRegionBodyStmt, this](InsertPointTy AllocaIP,
-                                                   InsertPointTy CodeGenIP) {
+    auto BodyGenCB = [SectionRegionBodyStmt,
+                      this](InsertPointTy AllocIP, InsertPointTy CodeGenIP,
+                            ArrayRef<InsertPointTy> DeallocIPs) {
       OMPBuilderCBHelpers::EmitOMPInlinedRegionBody(
-          *this, SectionRegionBodyStmt, AllocaIP, CodeGenIP, "section");
+          *this, SectionRegionBodyStmt, AllocIP, CodeGenIP, "section");
       return llvm::Error::success();
     };
 
@@ -5089,10 +5093,11 @@ void CodeGenFunction::EmitOMPMasterDirective(const OMPMasterDirective &S) {
       return llvm::Error::success();
     };
 
-    auto BodyGenCB = [MasterRegionBodyStmt, this](InsertPointTy AllocaIP,
-                                                  InsertPointTy CodeGenIP) {
+    auto BodyGenCB = [MasterRegionBodyStmt,
+                      this](InsertPointTy AllocIP, InsertPointTy CodeGenIP,
+                            ArrayRef<InsertPointTy> DeallocIPs) {
       OMPBuilderCBHelpers::EmitOMPInlinedRegionBody(
-          *this, MasterRegionBodyStmt, AllocaIP, CodeGenIP, "master");
+          *this, MasterRegionBodyStmt, AllocIP, CodeGenIP, "master");
       return llvm::Error::success();
     };
 
@@ -5139,10 +5144,11 @@ void CodeGenFunction::EmitOMPMaskedDirective(const OMPMaskedDirective &S) {
       return llvm::Error::success();
     };
 
-    auto BodyGenCB = [MaskedRegionBodyStmt, this](InsertPointTy AllocaIP,
-                                                  InsertPointTy CodeGenIP) {
+    auto BodyGenCB = [MaskedRegionBodyStmt,
+                      this](InsertPointTy AllocIP, InsertPointTy CodeGenIP,
+                            ArrayRef<InsertPointTy> DeallocIPs) {
       OMPBuilderCBHelpers::EmitOMPInlinedRegionBody(
-          *this, MaskedRegionBodyStmt, AllocaIP, CodeGenIP, "masked");
+          *this, MaskedRegionBodyStmt, AllocIP, CodeGenIP, "masked");
       return llvm::Error::success();
     };
 
@@ -5182,10 +5188,11 @@ void CodeGenFunction::EmitOMPCriticalDirective(const OMPCriticalDirective &S) {
       return llvm::Error::success();
     };
 
-    auto BodyGenCB = [CriticalRegionBodyStmt, this](InsertPointTy AllocaIP,
-                                                    InsertPointTy CodeGenIP) {
+    auto BodyGenCB = [CriticalRegionBodyStmt,
+                      this](InsertPointTy AllocIP, InsertPointTy CodeGenIP,
+                            ArrayRef<InsertPointTy> DeallocIPs) {
       OMPBuilderCBHelpers::EmitOMPInlinedRegionBody(
-          *this, CriticalRegionBodyStmt, AllocaIP, CodeGenIP, "critical");
+          *this, CriticalRegionBodyStmt, AllocIP, CodeGenIP, "critical");
       return llvm::Error::success();
     };
 
@@ -6152,8 +6159,8 @@ void CodeGenFunction::EmitOMPTaskgroupDirective(
     InsertPointTy AllocaIP(AllocaInsertPt->getParent(),
                            AllocaInsertPt->getIterator());
 
-    auto BodyGenCB = [&, this](InsertPointTy AllocaIP,
-                               InsertPointTy CodeGenIP) {
+    auto BodyGenCB = [&, this](InsertPointTy AllocIP, InsertPointTy CodeGenIP,
+                               ArrayRef<InsertPointTy> DeallocIPs) {
       Builder.restoreIP(CodeGenIP);
       EmitStmt(S.getInnermostCapturedStmt()->getCapturedStmt());
       return llvm::Error::success();
@@ -6162,7 +6169,8 @@ void CodeGenFunction::EmitOMPTaskgroupDirective(
     if (!CapturedStmtInfo)
       CapturedStmtInfo = &CapStmtInfo;
     llvm::OpenMPIRBuilder::InsertPointTy AfterIP =
-        cantFail(OMPBuilder.createTaskgroup(Builder, AllocaIP, BodyGenCB));
+        cantFail(OMPBuilder.createTaskgroup(Builder, AllocaIP,
+                                            /*DeallocIPs=*/{}, BodyGenCB));
     Builder.restoreIP(AfterIP);
     return;
   }
@@ -6879,8 +6887,9 @@ void CodeGenFunction::EmitOMPOrderedDirective(const OMPOrderedDirective &S) {
         return llvm::Error::success();
       };
 
-      auto BodyGenCB = [&S, C, this](InsertPointTy AllocaIP,
-                                     InsertPointTy CodeGenIP) {
+      auto BodyGenCB = [&S, C, this](InsertPointTy AllocIP,
+                                     InsertPointTy CodeGenIP,
+                                     ArrayRef<InsertPointTy> DeallocIPs) {
         Builder.restoreIP(CodeGenIP);
 
         const CapturedStmt *CS = S.getInnermostCapturedStmt();
@@ -6898,7 +6907,7 @@ void CodeGenFunction::EmitOMPOrderedDirective(const OMPOrderedDirective &S) {
                                                OutlinedFn, CapturedVars);
         } else {
           OMPBuilderCBHelpers::EmitOMPInlinedRegionBody(
-              *this, CS->getCapturedStmt(), AllocaIP, CodeGenIP, "ordered");
+              *this, CS->getCapturedStmt(), AllocIP, CodeGenIP, "ordered");
         }
         return llvm::Error::success();
       };

flang/include/flang/Optimizer/Support/InitFIR.h

@@ -34,6 +34,7 @@
 #include "mlir/Dialect/Math/IR/Math.h"
 #include "mlir/Dialect/OpenACC/OpenACC.h"
 #include "mlir/Dialect/OpenACC/Transforms/Passes.h"
+#include "mlir/Dialect/OpenMP/Transforms/Passes.h"
 #include "mlir/Dialect/SCF/IR/SCF.h"
 #include "mlir/Dialect/SCF/Transforms/Passes.h"
 #include "mlir/InitAllDialects.h"
@@ -106,6 +107,7 @@ inline void loadDialects(mlir::MLIRContext &context) {
 /// but is a smaller set since we aren't using many of the passes found there.
 inline void registerMLIRPassesForFortranTools() {
   mlir::acc::registerOpenACCPasses();
+  mlir::omp::registerOpenMPPasses();
   mlir::registerCanonicalizerPass();
   mlir::registerCSEPass();
   mlir::affine::registerAffineLoopFusionPass();

flang/lib/Optimizer/CodeGen/CodeGenOpenMP.cpp

@@ -245,8 +245,7 @@ struct TargetAllocMemOpConversion
       size = mlir::LLVM::MulOp::create(rewriter, loc, ity, size, scaleSize);
     for (mlir::Value opnd : adaptor.getOperands().drop_front())
       size = mlir::LLVM::MulOp::create(
-          rewriter, loc, ity, size,
-          integerCast(lowerTy(), loc, rewriter, ity, opnd));
+          rewriter, loc, ity, size, integerCast(lowerTy(), loc, rewriter, ity, opnd));
     auto mallocTyWidth = lowerTy().getIndexTypeBitwidth();
     auto mallocTy =
         mlir::IntegerType::get(rewriter.getContext(), mallocTyWidth);

flang/lib/Optimizer/Passes/Pipelines.cpp

@@ -10,6 +10,7 @@
 /// common to flang and the test tools.
 
 #include "flang/Optimizer/Passes/Pipelines.h"
+#include "mlir/Dialect/OpenMP/Transforms/Passes.h"
 #include "llvm/Support/CommandLine.h"
 
 /// Force setting the no-alias attribute on fuction arguments when possible.
@@ -408,6 +409,9 @@ void createDefaultFIRCodeGenPassPipeline(mlir::PassManager &pm,
   }
 
   fir::addFIRToLLVMPass(pm, config);
+
+  if (config.EnableOpenMP && !config.EnableOpenMPSimd)
+    pm.addPass(mlir::omp::createStackToSharedPass());
 }
 
 /// Create a pass pipeline for lowering from MLIR to LLVM IR

flang/test/Fir/basic-program.fir

@@ -161,5 +161,7 @@ func.func @_QQmain() {
 // PASSES-NEXT:  LowerNontemporalPass
 // PASSES-NEXT: FIRToLLVMLowering
 // PASSES-NEXT: ReconcileUnrealizedCasts
+// PASSES-NEXT: 'llvm.func' Pipeline
+// PASSES-NEXT:  StackToSharedPass
 // PASSES-NEXT: PrepareForOMPOffloadPrivatizationPass
 // PASSES-NEXT: LLVMIRLoweringPass

flang/test/Integration/OpenMP/threadprivate-target-device.f90

@@ -14,16 +14,14 @@
 ! target code in the same function.
 
 ! CHECK: define weak_odr protected amdgpu_kernel void @{{.*}}(ptr %{{.*}}, ptr %[[ARG1:.*]], ptr %[[ARG2:.*]]) #{{[0-9]+}} {
-! CHECK:  %[[ALLOCA_X:.*]] = alloca ptr, align 8, addrspace(5)
-! CHECK:  %[[ASCAST_X:.*]] = addrspacecast ptr addrspace(5) %[[ALLOCA_X]] to ptr
-! CHECK:  store ptr %[[ARG1]], ptr %[[ASCAST_X]], align 8
+! CHECK:  %[[ALLOC_N:.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 8)
+! CHECK:  store ptr %[[ARG2]], ptr %[[ALLOC_N]], align 8
 
-! CHECK:  %[[ALLOCA_N:.*]] = alloca ptr, align 8, addrspace(5)
-! CHECK:  %[[ASCAST_N:.*]] = addrspacecast ptr addrspace(5) %[[ALLOCA_N]] to ptr
-! CHECK:  store ptr %[[ARG2]], ptr %[[ASCAST_N]], align 8
+! CHECK:  %[[ALLOC_X:.*]] = call align 8 ptr @__kmpc_alloc_shared(i64 8)
+! CHECK:  store ptr %[[ARG1]], ptr %[[ALLOC_X]], align 8
 
-! CHECK:  %[[LOAD_X:.*]] = load ptr, ptr %[[ASCAST_X]], align 8
-! CHECK:  call void @bar_(ptr %[[LOAD_X]], ptr %[[ASCAST_N]])
+! CHECK:  %[[LOAD_X:.*]] = load ptr, ptr %[[ALLOC_X]], align 8
+! CHECK:  call void @bar_(ptr %[[LOAD_X]], ptr %[[ALLOC_N]])
 
 module test
   implicit none