[OPENMP] Support lastprivate clause in omp simd directive.

Added codegen for lastprivate clauses within simd loop-based directives.

llvm-svn: 239813

3 files changed, 41 insertions, 20 deletions

diff --git a/clang/lib/CodeGen/CGStmtOpenMP.cpp b/clang/lib/CodeGen/CGStmtOpenMP.cpp
index a6da3891af5..907fe93efc4 100644
--- a/clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ b/clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -316,26 +316,32 @@ void CodeGenFunction::EmitOMPLastprivateClauseFinal(
   //   ...
   //   orig_varn = private_orig_varn;
   // }
-  auto *ThenBB = createBasicBlock(".omp.lastprivate.then");
-  auto *DoneBB = createBasicBlock(".omp.lastprivate.done");
-  Builder.CreateCondBr(IsLastIterCond, ThenBB, DoneBB);
-  EmitBlock(ThenBB);
+  llvm::BasicBlock *ThenBB = nullptr;
+  llvm::BasicBlock *DoneBB = nullptr;
+  if (IsLastIterCond) {
+    ThenBB = createBasicBlock(".omp.lastprivate.then");
+    DoneBB = createBasicBlock(".omp.lastprivate.done");
+    Builder.CreateCondBr(IsLastIterCond, ThenBB, DoneBB);
+    EmitBlock(ThenBB);
+  }
   llvm::DenseMap<const Decl *, const Expr *> LoopCountersAndUpdates;
   const Expr *LastIterVal = nullptr;
   const Expr *IVExpr = nullptr;
   const Expr *IncExpr = nullptr;
   if (auto *LoopDirective = dyn_cast<OMPLoopDirective>(&D)) {
-    LastIterVal =
-        cast<VarDecl>(cast<DeclRefExpr>(LoopDirective->getUpperBoundVariable())
-                          ->getDecl())
-            ->getAnyInitializer();
-    IVExpr = LoopDirective->getIterationVariable();
-    IncExpr = LoopDirective->getInc();
-    auto IUpdate = LoopDirective->updates().begin();
-    for (auto *E : LoopDirective->counters()) {
-      auto *D = cast<DeclRefExpr>(E)->getDecl()->getCanonicalDecl();
-      LoopCountersAndUpdates[D] = *IUpdate;
-      ++IUpdate;
+    if (isOpenMPWorksharingDirective(D.getDirectiveKind())) {
+      LastIterVal = cast<VarDecl>(cast<DeclRefExpr>(
+                                      LoopDirective->getUpperBoundVariable())
+                                      ->getDecl())
+                        ->getAnyInitializer();
+      IVExpr = LoopDirective->getIterationVariable();
+      IncExpr = LoopDirective->getInc();
+      auto IUpdate = LoopDirective->updates().begin();
+      for (auto *E : LoopDirective->counters()) {
+        auto *D = cast<DeclRefExpr>(E)->getDecl()->getCanonicalDecl();
+        LoopCountersAndUpdates[D] = *IUpdate;
+        ++IUpdate;
+      }
     }
   }
   {
@@ -355,7 +361,7 @@ void CodeGenFunction::EmitOMPLastprivateClauseFinal(
           // directive, update its value before copyin back to original
           // variable.
           if (auto *UpExpr = LoopCountersAndUpdates.lookup(CanonicalVD)) {
-            if (FirstLCV) {
+            if (FirstLCV && LastIterVal) {
               EmitAnyExprToMem(LastIterVal, EmitLValue(IVExpr).getAddress(),
                                IVExpr->getType().getQualifiers(),
                                /*IsInitializer=*/false);
@@ -379,7 +385,9 @@ void CodeGenFunction::EmitOMPLastprivateClauseFinal(
       }
     }
   }
-  EmitBlock(DoneBB, /*IsFinished=*/true);
+  if (IsLastIterCond) {
+    EmitBlock(DoneBB, /*IsFinished=*/true);
+  }
 }
 
 void CodeGenFunction::EmitOMPReductionClauseInit(
@@ -793,11 +801,13 @@ void CodeGenFunction::EmitOMPSimdDirective(const OMPSimdDirective &S) {
         }
     }
 
+    bool HasLastprivateClause;
     {
       OMPPrivateScope LoopScope(CGF);
       EmitPrivateLoopCounters(CGF, LoopScope, S.counters());
       EmitPrivateLinearVars(CGF, S, LoopScope);
       CGF.EmitOMPPrivateClause(S, LoopScope);
+      HasLastprivateClause = CGF.EmitOMPLastprivateClauseInit(S, LoopScope);
       (void)LoopScope.Privatize();
       CGF.EmitOMPInnerLoop(S, LoopScope.requiresCleanups(),
                            S.getCond(), S.getInc(),
@@ -806,6 +816,10 @@ void CodeGenFunction::EmitOMPSimdDirective(const OMPSimdDirective &S) {
                              CGF.EmitStopPoint(&S);
                            },
                            [](CodeGenFunction &) {});
+      // Emit final copy of the lastprivate variables at the end of loops.
+      if (HasLastprivateClause) {
+        CGF.EmitOMPLastprivateClauseFinal(S);
+      }
     }
     CGF.EmitOMPSimdFinal(S);
     // Emit: if (PreCond) - end.
diff --git a/clang/lib/CodeGen/CodeGenFunction.h b/clang/lib/CodeGen/CodeGenFunction.h
index b3c31f61d1a..fddecbc9394 100644
--- a/clang/lib/CodeGen/CodeGenFunction.h
+++ b/clang/lib/CodeGen/CodeGenFunction.h
@@ -2130,9 +2130,9 @@ public:
   /// \param D Directive that has at least one 'lastprivate' directives.
   /// \param IsLastIterCond Boolean condition that must be set to 'i1 true' if
   /// it is the last iteration of the loop code in associated directive, or to
-  /// 'i1 false' otherwise.
+  /// 'i1 false' otherwise. If this item is nullptr, no final check is required.
   void EmitOMPLastprivateClauseFinal(const OMPExecutableDirective &D,
-                                     llvm::Value *IsLastIterCond);
+                                     llvm::Value *IsLastIterCond = nullptr);
   /// \brief Emit initial code for reduction variables. Creates reduction copies
   /// and initializes them with the values according to OpenMP standard.
   ///
diff --git a/clang/test/OpenMP/simd_codegen.cpp b/clang/test/OpenMP/simd_codegen.cpp
index 6b3170d5dfd..0a5b38a0ce3 100644
--- a/clang/test/OpenMP/simd_codegen.cpp
+++ b/clang/test/OpenMP/simd_codegen.cpp
@@ -182,6 +182,8 @@ void simple(float *a, float *b, float *c, float *d) {
   }
 
   int A;
+  // CHECK: store i32 -1, i32* [[A:%.+]],
+  A = -1;
   #pragma omp simd lastprivate(A)
 // Clause 'lastprivate' implementation is not completed yet.
 // Test checks that one iteration is separated in presence of lastprivate.
@@ -198,13 +200,18 @@ void simple(float *a, float *b, float *c, float *d) {
 // CHECK: [[IV7_0:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID]]
 // CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV7_0]], 3
 // CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
-// CHECK-NEXT: store i64 [[LC_IT_2]], i64* {{.+}}, !llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID]]
+// CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID]]
+// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID]]
+// CHECK-NEXT: [[CONV:%.+]] = trunc i64 [[LC_VAL]] to i32
+// CHECK-NEXT: store i32 [[CONV]], i32* [[A_PRIV:%[^,]+]],{{.+}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID]]
     A = i;
 // CHECK: [[IV7_2:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID]]
 // CHECK-NEXT: [[ADD7_2:%.+]] = add nsw i64 [[IV7_2]], 1
 // CHECK-NEXT: store i64 [[ADD7_2]], i64* [[OMP_IV7]]{{.*}}!llvm.mem.parallel_loop_access ![[SIMPLE_LOOP7_ID]]
   }
 // CHECK: [[SIMPLE_LOOP7_END]]
+// CHECK-NEXT: [[A_PRIV_VAL:%.+]] = load i32, i32* [[A_PRIV]],
+// CHECK-NEXT: store i32 [[A_PRIV_VAL]], i32* [[A]],
 // CHECK-NEXT: ret void
 }