[Flang][OpenMP][MLIR] Lowering of reduction,inreduction, nogroup and lastprivate clause to MLIR

flang/lib/Lower/OpenMP/ClauseProcessor.cpp

@@ -406,6 +406,11 @@ bool ClauseProcessor::processMergeable(
   return markClauseOccurrence<omp::clause::Mergeable>(result.mergeable);
 }
 
+bool ClauseProcessor::processNogroup(
+    mlir::omp::NogroupClauseOps &result) const {
+  return markClauseOccurrence<omp::clause::Nogroup>(result.nogroup);
+}
+
 bool ClauseProcessor::processNowait(mlir::omp::NowaitClauseOps &result) const {
   return markClauseOccurrence<omp::clause::Nowait>(result.nowait);
 }

flang/lib/Lower/OpenMP/ClauseProcessor.h

@@ -89,6 +89,7 @@ class ClauseProcessor {
   bool processInclusive(mlir::Location currentLocation,
                         mlir::omp::InclusiveClauseOps &result) const;
   bool processMergeable(mlir::omp::MergeableClauseOps &result) const;
+  bool processNogroup(mlir::omp::NogroupClauseOps &result) const;
   bool processNowait(mlir::omp::NowaitClauseOps &result) const;
   bool processNumTasks(lower::StatementContext &stmtCtx,
                        mlir::omp::NumTasksClauseOps &result) const;

flang/lib/Lower/OpenMP/DataSharingProcessor.cpp

@@ -342,7 +342,8 @@ void DataSharingProcessor::insertLastPrivateCompare(mlir::Operation *op) {
   if (!hasLastPrivate)
     return;
 
-  if (mlir::isa<mlir::omp::WsloopOp>(op) || mlir::isa<mlir::omp::SimdOp>(op)) {
+  if (mlir::isa<mlir::omp::WsloopOp>(op) || mlir::isa<mlir::omp::SimdOp>(op) ||
+      mlir::isa<mlir::omp::TaskloopOp>(op)) {
     mlir::omp::LoopRelatedClauseOps result;
     llvm::SmallVector<const semantics::Symbol *> iv;
     collectLoopRelatedInfo(converter, converter.getCurrentLocation(), eval,
@@ -408,7 +409,7 @@ void DataSharingProcessor::insertLastPrivateCompare(mlir::Operation *op) {
   } else {
     TODO(converter.getCurrentLocation(),
          "lastprivate clause in constructs other than "
-         "simd/worksharing-loop");
+         "simd/worksharing-loop/taskloop");
   }
 }
 

flang/lib/Lower/OpenMP/OpenMP.cpp

@@ -1763,21 +1763,24 @@ static void genTaskgroupClauses(
   cp.processTaskReduction(loc, clauseOps, taskReductionSyms);
 }
 
-static void genTaskloopClauses(lower::AbstractConverter &converter,
-                               semantics::SemanticsContext &semaCtx,
-                               lower::StatementContext &stmtCtx,
-                               const List<Clause> &clauses, mlir::Location loc,
-                               mlir::omp::TaskloopOperands &clauseOps) {
+static void genTaskloopClauses(
+    lower::AbstractConverter &converter, semantics::SemanticsContext &semaCtx,
+    lower::StatementContext &stmtCtx, const List<Clause> &clauses,
+    mlir::Location loc, mlir::omp::TaskloopOperands &clauseOps,
+    llvm::SmallVectorImpl<const semantics::Symbol *> &reductionSyms,
+    llvm::SmallVectorImpl<const semantics::Symbol *> &inReductionSyms) {
 
   ClauseProcessor cp(converter, semaCtx, clauses);
   cp.processGrainsize(stmtCtx, clauseOps);
   cp.processNumTasks(stmtCtx, clauseOps);
+  cp.processNogroup(clauseOps);
+  cp.processReduction(loc, clauseOps, reductionSyms);
+  cp.processInReduction(loc, clauseOps, inReductionSyms);
 
   cp.processTODO<clause::Allocate, clause::Collapse, clause::Default,
-                 clause::Final, clause::If, clause::InReduction,
-                 clause::Lastprivate, clause::Mergeable, clause::Nogroup,
-                 clause::Priority, clause::Reduction, clause::Shared,
-                 clause::Untied>(loc, llvm::omp::Directive::OMPD_taskloop);
+                 clause::Final, clause::If, clause::Mergeable, clause::Priority,
+                 clause::Shared, clause::Untied>(
+      loc, llvm::omp::Directive::OMPD_taskloop);
 }
 
 static void genTaskwaitClauses(lower::AbstractConverter &converter,
@@ -2979,8 +2982,11 @@ static mlir::omp::TaskloopOp genStandaloneTaskloop(
     lower::pft::Evaluation &eval, mlir::Location loc,
     const ConstructQueue &queue, ConstructQueue::const_iterator item) {
   mlir::omp::TaskloopOperands taskloopClauseOps;
+  llvm::SmallVector<const semantics::Symbol *> reductionSyms;
+  llvm::SmallVector<const semantics::Symbol *> inReductionSyms;
+
   genTaskloopClauses(converter, semaCtx, stmtCtx, item->clauses, loc,
-                     taskloopClauseOps);
+                     taskloopClauseOps, reductionSyms, inReductionSyms);
   DataSharingProcessor dsp(converter, semaCtx, item->clauses, eval,
                            /*shouldCollectPreDeterminedSymbols=*/true,
                            enableDelayedPrivatization, symTable);
@@ -2994,6 +3000,10 @@ static mlir::omp::TaskloopOp genStandaloneTaskloop(
   EntryBlockArgs taskloopArgs;
   taskloopArgs.priv.syms = dsp.getDelayedPrivSymbols();
   taskloopArgs.priv.vars = taskloopClauseOps.privateVars;
+  taskloopArgs.reduction.syms = reductionSyms;
+  taskloopArgs.reduction.vars = taskloopClauseOps.reductionVars;
+  taskloopArgs.inReduction.syms = inReductionSyms;
+  taskloopArgs.inReduction.vars = taskloopClauseOps.inReductionVars;
 
   auto taskLoopOp = genWrapperOp<mlir::omp::TaskloopOp>(
       converter, loc, taskloopClauseOps, taskloopArgs);

flang/test/Lower/OpenMP/taskloop.f90

@@ -1,6 +1,15 @@
 ! RUN: bbc -emit-hlfir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
 ! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
 
+! CHECK-LABEL:  omp.private
+! CHECK-SAME:       {type = private} @[[LAST_PRIVATE_I:.*]] : i32
+
+! CHECK-LABEL:  omp.private
+! CHECK-SAME:       {type = private} @[[LAST_PRIVATE_X:.*]] : i32
+
+! CHECK-LABEL:  omp.private
+! CHECK-SAME:       {type = private} @[[QFOMP_TASKLOOP_NOGROUPEI_PRIVATE_I32:.*]] : i32
+
 ! CHECK-LABEL:  omp.private 
 ! CHECK-SAME:       {type = private} @[[I_PRIVATE_TEST2:.*]] : i32
 
@@ -70,3 +79,46 @@ subroutine omp_taskloop_private
 ! CHECK:         }
   !$omp end taskloop
 end subroutine omp_taskloop_private
+
+!===============================================================================
+! `nogroup` clause
+!===============================================================================
+
+subroutine omp_taskloop_nogroup()
+  ! CHECK: omp.taskloop nogroup
+  !$omp taskloop nogroup
+  do i = 1, 10
+    call foo()
+  end do
+  !$omp end taskloop
+end subroutine
+
+!===============================================================================
+! `lastprivate` clause
+!===============================================================================
+
+! CHECK-LABEL:  func.func @_QPomp_taskloop_lastprivate
+! CHECK:          %[[ALLOCA_I:.*]] = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFomp_taskloop_lastprivateEi"}
+! CHECK:          %[[DECL_I:.*]]:2 = hlfir.declare %[[ALLOCA_I]] {uniq_name = "_QFomp_taskloop_lastprivateEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK:          %[[ALLOCA_X:.*]] = fir.alloca i32 {bindc_name = "x", uniq_name = "_QFomp_taskloop_lastprivateEx"}
+! CHECK:          %[[DECL_X:.*]]:2 = hlfir.declare %[[ALLOCA_X]] {uniq_name = "_QFomp_taskloop_lastprivateEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+subroutine omp_taskloop_lastprivate()
+   integer x
+   x = 0
+   ! CHECK:  omp.taskloop private(@[[LAST_PRIVATE_X]] %[[DECL_X]]#0 -> %[[ARG0]], @[[LAST_PRIVATE_I]] %[[DECL_I]]#0 -> %[[ARG1]] : !fir.ref<i32>, !fir.ref<i32>) {
+   !$omp taskloop lastprivate(x)
+   do i = 1, 100
+      ! CHECK: %[[DECL_ARG0:.*]]:2 = hlfir.declare %[[ARG0]] {uniq_name = "_QFomp_taskloop_lastprivateEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+      ! CHECK: %[[LOAD_ARG0:.*]] = fir.load %[[DECL_ARG0]]#0 : !fir.ref<i32>
+      ! CHECK: %[[RES_ADD:.*]] = arith.addi %[[LOAD_ARG0]], %{{.*}} : i32
+      ! CHECK:  hlfir.assign %[[RES_ADD]] to %[[DECL_ARG0]]#0 : i32, !fir.ref<i32>
+      x = x + 1
+      ! CHECK:  %[[SELCT_RESULT:.*]] = arith.select %{{.*}}, %{{.*}}, %{{.*}} : i1
+      ! CHECK:  fir.if %[[SELCT_RESULT]] {
+      ! CHECK:    %[[LOADED_SUM:.*]] = fir.load %[[DECL_ARG0]]#0 : !fir.ref<i32>
+      ! CHECK:    hlfir.assign %[[LOADED_SUM]] to %[[DECL_X]]#0 : i32, !fir.ref<i32>
+      ! CHECK:  }
+      ! CHECK:  omp.yield
+   end do
+   !$omp end taskloop
+end subroutine omp_taskloop_lastprivate

flang/test/Lower/taskloop-inreduction.f90

@@ -0,0 +1,40 @@
+! RUN: bbc -emit-hlfir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
+
+! CHECK-LABEL:  omp.private
+! CHECK-SAME:        {type = private} @[[PRIVATE_I:.*]] : i32
+
+! CHECK-LABEL:  omp.declare_reduction 
+! CHECK-SAME:   @[[ADD_RED_I32:.*]] : i32 init {
+! CHECK:       ^bb0(%{{.*}}: i32):
+! CHECK:        %[[C0_I32:.*]] = arith.constant 0 : i32
+! CHECK:        omp.yield(%[[C0_I32]] : i32)
+! CHECK:     } combiner {
+! CHECK:     ^bb0(%{{.*}}: i32, %{{.*}}: i32):
+! CHECK:        %[[RES:.*]] = arith.addi %{{.*}}, %{{.*}} : i32
+! CHECK:        omp.yield(%[[RES]] : i32)
+! CHECK:     }
+
+! CHECK-LABEL: func.func @_QPomp_taskloop_inreduction
+! CHECK:          %[[ALLOCA_I:.*]] = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFomp_taskloop_inreductionEi"}
+! CHECK:          %[[DECL_I:.*]]:2 = hlfir.declare %[[ALLOCA_I]] {uniq_name = "_QFomp_taskloop_inreductionEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK:          %[[ALLOCA_X:.*]] = fir.alloca i32 {bindc_name = "x", uniq_name = "_QFomp_taskloop_inreductionEx"}
+! CHECK:          %[[DECL_X:.*]]:2 = hlfir.declare %[[ALLOCA_X]] {uniq_name = "_QFomp_taskloop_inreductionEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK:          %[[INIT_X:.*]] = arith.constant 0 : i32
+! CHECK:          hlfir.assign %[[INIT_X]] to %[[DECL_X]]#0 : i32, !fir.ref<i32>
+subroutine omp_taskloop_inreduction()
+   integer x
+   x = 0
+   ! CHECK:        omp.taskloop in_reduction(@[[ADD_RED_I32]] 
+   ! CHECK:        %[[DECL_X]]#0 -> %[[ARG0:.*]] : !fir.ref<i32>) private(@[[PRIVATE_I]] %[[DECL_I]]#0 -> %[[ARG1:.*]] : !fir.ref<i32>) {
+   ! CHECK:        %[[VAL_ARG1:.*]]:2 = hlfir.declare %[[ARG0]] 
+   ! CHECK-SAME:   {uniq_name = "_QFomp_taskloop_inreductionEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+   !$omp taskloop in_reduction(+:x)
+   do i = 1, 100
+      ! CHECK: %[[X_VAL:.*]] = fir.load %[[VAL_ARG1]]#0 : !fir.ref<i32>
+      ! CHECK: %[[ADD_VAL:.*]] = arith.addi %[[X_VAL]], %{{.*}} : i32
+      x = x + 1
+      ! CHECK: hlfir.assign %[[ADD_VAL]] to %[[VAL_ARG1]]#0 : i32, !fir.ref<i32>
+   end do
+   !$omp end taskloop
+end subroutine omp_taskloop_inreduction

flang/test/Lower/taskloop-reduction.f90

@@ -0,0 +1,39 @@
+! RUN: bbc -emit-hlfir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
+
+! CHECK-LABEL:  omp.private
+! CHECK-SAME:        {type = private} @[[PRIVATE_I:.*]] : i32
+
+! CHECK-LABEL:  omp.declare_reduction 
+! CHECK-SAME:   @[[ADD_RED_I32:.*]] : i32 init {
+! CHECK:       ^bb0(%{{.*}}: i32):
+! CHECK:        %[[C0_I32:.*]] = arith.constant 0 : i32
+! CHECK:        omp.yield(%[[C0_I32]] : i32)
+! CHECK:     } combiner {
+! CHECK:     ^bb0(%{{.*}}: i32, %{{.*}}: i32):
+! CHECK:        %[[RES:.*]] = arith.addi %{{.*}}, %{{.*}} : i32
+! CHECK:        omp.yield(%[[RES]] : i32)
+! CHECK:     }
+
+! CHECK-LABEL:  func.func @_QPomp_taskloop_reduction
+! CHECK:          %[[ALLOCA_I:.*]] = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFomp_taskloop_reductionEi"}
+! CHECK:          %[[DECL_I:.*]]:2 = hlfir.declare %[[ALLOCA_I]] {uniq_name = "_QFomp_taskloop_reductionEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK:          %[[ALLOCA_X:.*]] = fir.alloca i32 {bindc_name = "x", uniq_name = "_QFomp_taskloop_reductionEx"}
+! CHECK:          %[[DECL_X:.*]]:2 = hlfir.declare %[[ALLOCA_X]] {uniq_name = "_QFomp_taskloop_reductionEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK:          %[[INIT_X:.*]] = arith.constant 0 : i32
+! CHECK:          hlfir.assign %[[INIT_X]] to %[[DECL_X]]#0 : i32, !fir.ref<i32>
+subroutine omp_taskloop_reduction()
+   integer x
+   x = 0
+   ! CHECK:       omp.taskloop private(@[[PRIVATE_I]] 
+   ! CHECK-SAME:  %[[DECL_I]]#0 -> %[[ARG0:.*]] : !fir.ref<i32>) reduction(@[[ADD_RED_I32]] %[[DECL_X]]#0 -> %[[ARG1:.*]] : !fir.ref<i32>) {
+   ! CHECK:       %[[VAL_ARG1:.*]]:2 = hlfir.declare %[[ARG1]] 
+   !$omp taskloop reduction(+:x)
+   do i = 1, 100
+      ! CHECK: %[[X_VAL:.*]] = fir.load %[[VAL_ARG1]]#0 : !fir.ref<i32>
+      ! CHECK: %[[ADD_VAL:.*]] = arith.addi %[[X_VAL]], %{{.*}} : i32
+      x = x + 1
+      ! CHECK: hlfir.assign %[[ADD_VAL]] to %[[VAL_ARG1]]#0 : i32, !fir.ref<i32>
+   end do
+   !$omp end taskloop
+end subroutine omp_taskloop_reduction