[flang-commits] [flang] [Flang][OpenMP]Support for lowering task_reduction and in_reduction to MLIR (PR #111155)

Kaviya Rajendiran via flang-commits flang-commits at lists.llvm.org
Mon Oct 7 00:54:27 PDT 2024


https://github.com/kaviya2510 updated https://github.com/llvm/llvm-project/pull/111155

>From 711fe3367b6e1041dc2e45d1d4e509ccdfa8b101 Mon Sep 17 00:00:00 2001
From: Kaviya Rajendiran <kaviyara2000 at gmail.com>
Date: Mon, 7 Oct 2024 13:24:02 +0530
Subject: [PATCH] [Flang] [OpenMP] Support for lowering task_reduction and
 in_reduction to MLIR

---
 flang/lib/Lower/OpenMP/ClauseProcessor.cpp    | 76 ++++++++++++++++++-
 flang/lib/Lower/OpenMP/ClauseProcessor.h      | 10 +++
 flang/lib/Lower/OpenMP/OpenMP.cpp             | 63 ++++++++++-----
 flang/lib/Lower/OpenMP/ReductionProcessor.cpp | 41 ++++++++--
 flang/lib/Lower/OpenMP/ReductionProcessor.h   |  4 +-
 .../Lower/OpenMP/task_array_reduction.f90     | 50 ++++++++++++
 flang/test/Lower/OpenMP/task_in_reduction.f90 | 48 ++++++++++++
 flang/test/Lower/OpenMP/task_reduction.f90    | 43 +++++++++++
 8 files changed, 307 insertions(+), 28 deletions(-)
 create mode 100644 flang/test/Lower/OpenMP/task_array_reduction.f90
 create mode 100644 flang/test/Lower/OpenMP/task_in_reduction.f90
 create mode 100644 flang/test/Lower/OpenMP/task_reduction.f90

diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index a4d2524bccf5c3..6a97b32d80c4b3 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -1063,7 +1063,7 @@ bool ClauseProcessor::processReduction(
         llvm::SmallVector<mlir::Attribute> reductionDeclSymbols;
         llvm::SmallVector<const semantics::Symbol *> reductionSyms;
         ReductionProcessor rp;
-        rp.addDeclareReduction(
+        rp.addDeclareReduction<omp::clause::Reduction>(
             currentLocation, converter, clause, reductionVars, reduceVarByRef,
             reductionDeclSymbols, outReductionSyms ? &reductionSyms : nullptr);
 
@@ -1085,6 +1085,80 @@ bool ClauseProcessor::processReduction(
       });
 }
 
+bool ClauseProcessor::processTaskReduction(
+    mlir::Location currentLocation, mlir::omp::TaskReductionClauseOps &result,
+    llvm::SmallVectorImpl<mlir::Type> *outReductionTypes,
+    llvm::SmallVectorImpl<const semantics::Symbol *> *outReductionSyms) const {
+  return findRepeatableClause<omp::clause::TaskReduction>(
+      [&](const omp::clause::TaskReduction &clause, const parser::CharBlock &) {
+        llvm::SmallVector<mlir::Value> taskReductionVars;
+        llvm::SmallVector<bool> taskReductionByref;
+        llvm::SmallVector<mlir::Attribute> taskReductionDeclSymbols;
+        llvm::SmallVector<const semantics::Symbol *> taskReductionSyms;
+        ReductionProcessor rp;
+        rp.addDeclareReduction<omp::clause::TaskReduction>(
+            currentLocation, converter, clause, taskReductionVars,
+            taskReductionByref, taskReductionDeclSymbols,
+            outReductionSyms ? &taskReductionSyms : nullptr);
+
+        // Copy local lists into the output.
+        llvm::copy(taskReductionVars,
+                   std::back_inserter(result.taskReductionVars));
+        llvm::copy(taskReductionByref,
+                   std::back_inserter(result.taskReductionByref));
+        llvm::copy(taskReductionDeclSymbols,
+                   std::back_inserter(result.taskReductionSyms));
+
+        if (outReductionTypes) {
+          outReductionTypes->reserve(outReductionTypes->size() +
+                                     taskReductionVars.size());
+          llvm::transform(taskReductionVars,
+                          std::back_inserter(*outReductionTypes),
+                          [](mlir::Value v) { return v.getType(); });
+        }
+
+        if (outReductionSyms)
+          llvm::copy(taskReductionSyms, std::back_inserter(*outReductionSyms));
+      });
+}
+
+bool ClauseProcessor::processInReduction(
+    mlir::Location currentLocation, mlir::omp::InReductionClauseOps &result,
+    llvm::SmallVectorImpl<mlir::Type> *outReductionTypes,
+    llvm::SmallVectorImpl<const semantics::Symbol *> *outReductionSyms) const {
+  return findRepeatableClause<omp::clause::InReduction>(
+      [&](const omp::clause::InReduction &clause,
+          const parser::CharBlock &source) {
+        llvm::SmallVector<mlir::Value> inReductionVars;
+        llvm::SmallVector<bool> inReductionByref;
+        llvm::SmallVector<mlir::Attribute> inReductionDeclSymbols;
+        llvm::SmallVector<const semantics::Symbol *> inReductionSyms;
+        ReductionProcessor rp;
+        rp.addDeclareReduction<omp::clause::InReduction>(
+            currentLocation, converter, clause, inReductionVars,
+            inReductionByref, inReductionDeclSymbols,
+            outReductionSyms ? &inReductionSyms : nullptr);
+
+        // Copy local lists into the output.
+        llvm::copy(inReductionVars, std::back_inserter(result.inReductionVars));
+        llvm::copy(inReductionByref,
+                   std::back_inserter(result.inReductionByref));
+        llvm::copy(inReductionDeclSymbols,
+                   std::back_inserter(result.inReductionSyms));
+
+        if (outReductionTypes) {
+          outReductionTypes->reserve(outReductionTypes->size() +
+                                     inReductionVars.size());
+          llvm::transform(inReductionVars,
+                          std::back_inserter(*outReductionTypes),
+                          [](mlir::Value v) { return v.getType(); });
+        }
+
+        if (outReductionSyms)
+          llvm::copy(inReductionSyms, std::back_inserter(*outReductionSyms));
+      });
+}
+
 bool ClauseProcessor::processTo(
     llvm::SmallVectorImpl<DeclareTargetCapturePair> &result) const {
   return findRepeatableClause<omp::clause::To>(
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.h b/flang/lib/Lower/OpenMP/ClauseProcessor.h
index 0c8e7bd47ab5a6..04416a927a1c37 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.h
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.h
@@ -129,6 +129,16 @@ class ClauseProcessor {
       llvm::SmallVectorImpl<mlir::Type> *reductionTypes = nullptr,
       llvm::SmallVectorImpl<const semantics::Symbol *> *reductionSyms =
           nullptr) const;
+  bool processTaskReduction(
+      mlir::Location currentLocation, mlir::omp::TaskReductionClauseOps &result,
+      llvm::SmallVectorImpl<mlir::Type> *taskReductionTypes = nullptr,
+      llvm::SmallVectorImpl<const semantics::Symbol *> *taskReductionSyms =
+          nullptr) const;
+  bool processInReduction(
+      mlir::Location currentLocation, mlir::omp::InReductionClauseOps &result,
+      llvm::SmallVectorImpl<mlir::Type> *inReductionTypes = nullptr,
+      llvm::SmallVectorImpl<const semantics::Symbol *> *inReductionSyms =
+          nullptr) const;
   bool processTo(llvm::SmallVectorImpl<DeclareTargetCapturePair> &result) const;
   bool processUseDeviceAddr(
       lower::StatementContext &stmtCtx,
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 8195f4a897a90b..7e1e2afc760f0d 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -1243,33 +1243,37 @@ static void genTargetEnterExitUpdateDataClauses(
   cp.processNowait(clauseOps);
 }
 
-static void genTaskClauses(lower::AbstractConverter &converter,
-                           semantics::SemanticsContext &semaCtx,
-                           lower::StatementContext &stmtCtx,
-                           const List<Clause> &clauses, mlir::Location loc,
-                           mlir::omp::TaskOperands &clauseOps) {
+static void genTaskClauses(
+    lower::AbstractConverter &converter, semantics::SemanticsContext &semaCtx,
+    lower::StatementContext &stmtCtx, const List<Clause> &clauses,
+    mlir::Location loc, mlir::omp::TaskOperands &clauseOps,
+    llvm::SmallVectorImpl<mlir::Type> &inReductionTypes,
+    llvm::SmallVectorImpl<const semantics::Symbol *> &inReductionSyms) {
   ClauseProcessor cp(converter, semaCtx, clauses);
   cp.processAllocate(clauseOps);
   cp.processDepend(clauseOps);
   cp.processFinal(stmtCtx, clauseOps);
   cp.processIf(llvm::omp::Directive::OMPD_task, clauseOps);
+  cp.processInReduction(loc, clauseOps, &inReductionTypes, &inReductionSyms);
   cp.processMergeable(clauseOps);
   cp.processPriority(stmtCtx, clauseOps);
   cp.processUntied(clauseOps);
   // TODO Support delayed privatization.
 
-  cp.processTODO<clause::Affinity, clause::Detach, clause::InReduction>(
+  cp.processTODO<clause::Affinity, clause::Detach>(
       loc, llvm::omp::Directive::OMPD_task);
 }
 
-static void genTaskgroupClauses(lower::AbstractConverter &converter,
-                                semantics::SemanticsContext &semaCtx,
-                                const List<Clause> &clauses, mlir::Location loc,
-                                mlir::omp::TaskgroupOperands &clauseOps) {
+static void genTaskgroupClauses(
+    lower::AbstractConverter &converter, semantics::SemanticsContext &semaCtx,
+    const List<Clause> &clauses, mlir::Location loc,
+    mlir::omp::TaskgroupOperands &clauseOps,
+    llvm::SmallVectorImpl<mlir::Type> &taskReductionTypes,
+    llvm::SmallVectorImpl<const semantics::Symbol *> &taskReductionSyms) {
   ClauseProcessor cp(converter, semaCtx, clauses);
   cp.processAllocate(clauseOps);
-  cp.processTODO<clause::TaskReduction>(loc,
-                                        llvm::omp::Directive::OMPD_taskgroup);
+  cp.processTaskReduction(loc, clauseOps, &taskReductionTypes,
+                          &taskReductionSyms);
 }
 
 static void genTaskwaitClauses(lower::AbstractConverter &converter,
@@ -1869,13 +1873,26 @@ genTaskOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
           ConstructQueue::const_iterator item) {
   lower::StatementContext stmtCtx;
   mlir::omp::TaskOperands clauseOps;
-  genTaskClauses(converter, semaCtx, stmtCtx, item->clauses, loc, clauseOps);
+  llvm::SmallVector<mlir::Type> inReductionTypes;
+  llvm::SmallVector<const semantics::Symbol *> inreductionSyms;
+  genTaskClauses(converter, semaCtx, stmtCtx, item->clauses, loc, clauseOps,
+                 inReductionTypes, inreductionSyms);
 
-  return genOpWithBody<mlir::omp::TaskOp>(
+  auto reductionCallback = [&](mlir::Operation *op) {
+    genReductionVars(op, converter, loc, inreductionSyms, inReductionTypes);
+    return inreductionSyms;
+  };
+
+  auto taskOp = genOpWithBody<mlir::omp::TaskOp>(
       OpWithBodyGenInfo(converter, symTable, semaCtx, loc, eval,
                         llvm::omp::Directive::OMPD_task)
-          .setClauses(&item->clauses),
+          .setClauses(&item->clauses)
+          .setGenRegionEntryCb(reductionCallback),
       queue, item, clauseOps);
+  // Add reduction variables as arguments
+  llvm::SmallVector<mlir::Location> blockArgLocs(inReductionTypes.size(), loc);
+  taskOp->getRegion(0).addArguments(inReductionTypes, blockArgLocs);
+  return taskOp;
 }
 
 static mlir::omp::TaskgroupOp
@@ -1885,13 +1902,21 @@ genTaskgroupOp(lower::AbstractConverter &converter, lower::SymMap &symTable,
                const ConstructQueue &queue,
                ConstructQueue::const_iterator item) {
   mlir::omp::TaskgroupOperands clauseOps;
-  genTaskgroupClauses(converter, semaCtx, item->clauses, loc, clauseOps);
+  llvm::SmallVector<mlir::Type> taskReductionTypes;
+  llvm::SmallVector<const semantics::Symbol *> taskReductionSyms;
+  genTaskgroupClauses(converter, semaCtx, item->clauses, loc, clauseOps,
+                      taskReductionTypes, taskReductionSyms);
 
-  return genOpWithBody<mlir::omp::TaskgroupOp>(
+  auto taskgroupOp = genOpWithBody<mlir::omp::TaskgroupOp>(
       OpWithBodyGenInfo(converter, symTable, semaCtx, loc, eval,
                         llvm::omp::Directive::OMPD_taskgroup)
           .setClauses(&item->clauses),
       queue, item, clauseOps);
+
+  // Add reduction variables as arguments
+  llvm::SmallVector<mlir::Location> blockArgLocs(taskReductionSyms.size(), loc);
+  taskgroupOp->getRegion(0).addArguments(taskReductionTypes, blockArgLocs);
+  return taskgroupOp;
 }
 
 static mlir::omp::TaskwaitOp
@@ -2767,7 +2792,9 @@ static void genOMP(lower::AbstractConverter &converter, lower::SymMap &symTable,
         !std::holds_alternative<clause::ThreadLimit>(clause.u) &&
         !std::holds_alternative<clause::Threads>(clause.u) &&
         !std::holds_alternative<clause::UseDeviceAddr>(clause.u) &&
-        !std::holds_alternative<clause::UseDevicePtr>(clause.u)) {
+        !std::holds_alternative<clause::UseDevicePtr>(clause.u) &&
+        !std::holds_alternative<clause::TaskReduction>(clause.u) &&
+        !std::holds_alternative<clause::InReduction>(clause.u)) {
       TODO(clauseLocation, "OpenMP Block construct clause");
     }
   }
diff --git a/flang/lib/Lower/OpenMP/ReductionProcessor.cpp b/flang/lib/Lower/OpenMP/ReductionProcessor.cpp
index 9da15ba303a475..66853aa121ff00 100644
--- a/flang/lib/Lower/OpenMP/ReductionProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ReductionProcessor.cpp
@@ -24,6 +24,7 @@
 #include "flang/Parser/tools.h"
 #include "mlir/Dialect/OpenMP/OpenMPDialect.h"
 #include "llvm/Support/CommandLine.h"
+#include <type_traits>
 
 static llvm::cl::opt<bool> forceByrefReduction(
     "force-byref-reduction",
@@ -34,6 +35,32 @@ namespace Fortran {
 namespace lower {
 namespace omp {
 
+// explicit template declarations
+template void ReductionProcessor::addDeclareReduction<omp::clause::Reduction>(
+    mlir::Location currentLocation, lower::AbstractConverter &converter,
+    const omp::clause::Reduction &reduction,
+    llvm::SmallVectorImpl<mlir::Value> &reductionVars,
+    llvm::SmallVectorImpl<bool> &reduceVarByRef,
+    llvm::SmallVectorImpl<mlir::Attribute> &reductionDeclSymbols,
+    llvm::SmallVectorImpl<const semantics::Symbol *> *reductionSymbols);
+
+template void
+ReductionProcessor::addDeclareReduction<omp::clause::TaskReduction>(
+    mlir::Location currentLocation, lower::AbstractConverter &converter,
+    const omp::clause::TaskReduction &reduction,
+    llvm::SmallVectorImpl<mlir::Value> &reductionVars,
+    llvm::SmallVectorImpl<bool> &reduceVarByRef,
+    llvm::SmallVectorImpl<mlir::Attribute> &reductionDeclSymbols,
+    llvm::SmallVectorImpl<const semantics::Symbol *> *reductionSymbols);
+
+template void ReductionProcessor::addDeclareReduction<omp::clause::InReduction>(
+    mlir::Location currentLocation, lower::AbstractConverter &converter,
+    const omp::clause::InReduction &reduction,
+    llvm::SmallVectorImpl<mlir::Value> &reductionVars,
+    llvm::SmallVectorImpl<bool> &reduceVarByRef,
+    llvm::SmallVectorImpl<mlir::Attribute> &reductionDeclSymbols,
+    llvm::SmallVectorImpl<const semantics::Symbol *> *reductionSymbols);
+
 ReductionProcessor::ReductionIdentifier ReductionProcessor::getReductionType(
     const omp::clause::ProcedureDesignator &pd) {
   auto redType = llvm::StringSwitch<std::optional<ReductionIdentifier>>(
@@ -716,22 +743,22 @@ static bool doReductionByRef(mlir::Value reductionVar) {
   return false;
 }
 
+template <class T>
 void ReductionProcessor::addDeclareReduction(
     mlir::Location currentLocation, lower::AbstractConverter &converter,
-    const omp::clause::Reduction &reduction,
-    llvm::SmallVectorImpl<mlir::Value> &reductionVars,
+    const T &reduction, llvm::SmallVectorImpl<mlir::Value> &reductionVars,
     llvm::SmallVectorImpl<bool> &reduceVarByRef,
     llvm::SmallVectorImpl<mlir::Attribute> &reductionDeclSymbols,
     llvm::SmallVectorImpl<const semantics::Symbol *> *reductionSymbols) {
   fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
-
-  if (std::get<std::optional<omp::clause::Reduction::ReductionModifier>>(
-          reduction.t))
-    TODO(currentLocation, "Reduction modifiers are not supported");
+  if constexpr (std::is_same<T, omp::clause::Reduction>::value) {
+    if (std::get<std::optional<typename T::ReductionModifier>>(reduction.t))
+      TODO(currentLocation, "Reduction modifiers are not supported");
+  }
 
   mlir::omp::DeclareReductionOp decl;
   const auto &redOperatorList{
-      std::get<omp::clause::Reduction::ReductionIdentifiers>(reduction.t)};
+      std::get<typename T::ReductionIdentifiers>(reduction.t)};
   assert(redOperatorList.size() == 1 && "Expecting single operator");
   const auto &redOperator = redOperatorList.front();
   const auto &objectList{std::get<omp::ObjectList>(reduction.t)};
diff --git a/flang/lib/Lower/OpenMP/ReductionProcessor.h b/flang/lib/Lower/OpenMP/ReductionProcessor.h
index 0ed5782e5da1b7..d71fde93de33d0 100644
--- a/flang/lib/Lower/OpenMP/ReductionProcessor.h
+++ b/flang/lib/Lower/OpenMP/ReductionProcessor.h
@@ -120,10 +120,10 @@ class ReductionProcessor {
 
   /// Creates a reduction declaration and associates it with an OpenMP block
   /// directive.
+  template <class T>
   static void addDeclareReduction(
       mlir::Location currentLocation, lower::AbstractConverter &converter,
-      const omp::clause::Reduction &reduction,
-      llvm::SmallVectorImpl<mlir::Value> &reductionVars,
+      const T &reduction, llvm::SmallVectorImpl<mlir::Value> &reductionVars,
       llvm::SmallVectorImpl<bool> &reduceVarByRef,
       llvm::SmallVectorImpl<mlir::Attribute> &reductionDeclSymbols,
       llvm::SmallVectorImpl<const semantics::Symbol *> *reductionSymbols =
diff --git a/flang/test/Lower/OpenMP/task_array_reduction.f90 b/flang/test/Lower/OpenMP/task_array_reduction.f90
new file mode 100644
index 00000000000000..74693343744c26
--- /dev/null
+++ b/flang/test/Lower/OpenMP/task_array_reduction.f90
@@ -0,0 +1,50 @@
+! 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.declare_reduction @add_reduction_byref_box_Uxf32 : !fir.ref<!fir.box<!fir.array<?xf32>>> alloc {
+! [...]
+! CHECK:  omp.yield
+! CHECK-LABEL:  } init {
+! [...]
+! CHECK:  omp.yield
+! CHECK-LABEL:  } combiner {
+! [...]
+! CHECK:  omp.yield
+! CHECK-LABEL:  }  cleanup {
+! [...]
+! CHECK:  omp.yield
+! CHECK:  }
+
+! CHECK-LABEL:  func.func @_QPtaskreduction
+! CHECK-SAME:  (%[[VAL_0:.*]]: !fir.box<!fir.array<?xf32>> {fir.bindc_name = "x"}) {
+! CHECK:  %[[VAL_1:.*]] = fir.dummy_scope : !fir.dscope
+! CHECK:  %[[VAL_2:.*]]:2 = hlfir.declare %[[VAL_0]] dummy_scope %[[VAL_1]]
+! CHECK-SAME  {uniq_name = "_QFtaskreductionEx"} : (!fir.box<!fir.array<?xf32>>, !fir.dscope) -> (!fir.box<!fir.array<?xf32>>, !fir.box<!fir.array<?xf32>>)
+! CHECK:  omp.parallel {
+! CHECK:  %[[VAL_3:.*]] = fir.alloca !fir.box<!fir.array<?xf32>>
+! CHECK:  fir.store %[[VAL_2]]#1 to %[[VAL_3]] : !fir.ref<!fir.box<!fir.array<?xf32>>>
+! CHECK:  omp.taskgroup task_reduction(byref @add_reduction_byref_box_Uxf32 %[[VAL_3]] ->  %[[VAL_4:.*]]: !fir.ref<!fir.box<!fir.array<?xf32>>>) {
+! CHECK:  %[[VAL_5:.*]] = fir.alloca !fir.box<!fir.array<?xf32>>
+! CHECK:  fir.store %[[VAL_2]]#1 to %[[VAL_5]] : !fir.ref<!fir.box<!fir.array<?xf32>>>
+! CHECK:  omp.task in_reduction(byref @add_reduction_byref_box_Uxf32 %[[VAL_5]] -> %[[VAL_6:.*]] : !fir.ref<!fir.box<!fir.array<?xf32>>>) {
+! [...]
+! CHECK:  omp.terminator
+! CHECK:  }
+! CHECK:  omp.terminator
+! CHECK:  }
+! CHECK:  omp.terminator
+! CHECK:  }
+! CHECK:  return
+! CHECK:  }
+
+subroutine taskReduction(x)
+   real, dimension(:) :: x
+   !$omp parallel
+   !$omp taskgroup task_reduction(+:x)
+   !$omp task in_reduction(+:x)
+   x = x + 1
+   !$omp end task
+   !$omp end taskgroup
+   !$omp end parallel
+end subroutine
+
diff --git a/flang/test/Lower/OpenMP/task_in_reduction.f90 b/flang/test/Lower/OpenMP/task_in_reduction.f90
new file mode 100644
index 00000000000000..26c079d5ac8aa5
--- /dev/null
+++ b/flang/test/Lower/OpenMP/task_in_reduction.f90
@@ -0,0 +1,48 @@
+! 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.declare_reduction
+!CHECK-SAME: @[[RED_I32_NAME:.*]] : i32 init {
+!CHECK: ^bb0(%{{.*}}: i32):
+!CHECK:  %[[C0_1:.*]] = arith.constant 0 : i32
+!CHECK:  omp.yield(%[[C0_1]] : i32)
+!CHECK: } combiner {
+!CHECK: ^bb0(%[[ARG0:.*]]: i32, %[[ARG1:.*]]: i32):
+!CHECK:  %[[RES:.*]] = arith.addi %[[ARG0]], %[[ARG1]] : i32
+!CHECK:  omp.yield(%[[RES]] : i32)
+!CHECK: }
+
+!CHECK-LABEL:  func.func @_QPin_reduction() {
+!CHECK:  %[[VAL_0:.*]] = fir.alloca i32 {bindc_name = "x", uniq_name = "_QFin_reductionEx"}
+!CHECK:  %[[VAL_1:.*]]:2 = hlfir.declare %[[VAL_0]] {uniq_name = "_QFin_reductionEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+!CHECK:  %[[VAL_2:.*]] = arith.constant 0 : i32
+!CHECK:  hlfir.assign %[[VAL_2]] to %[[VAL_1]]#0 : i32, !fir.ref<i32>
+!CHECK:  omp.parallel {
+!CHECK:  omp.taskgroup task_reduction(@[[RED_I32_NAME]] %[[VAL_1]]#0 -> %[[VAL_3:.*]] : !fir.ref<i32>) {
+!CHECK:  omp.task in_reduction(@[[RED_I32_NAME]] %[[VAL_1]]#0 -> %[[VAL_4:.*]] : !fir.ref<i32>) {
+!CHECK:  %[[VAL_5:.*]]:2 = hlfir.declare %[[VAL_4]] {uniq_name = "_QFin_reductionEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+!CHECK:  %[[VAL_6:.*]] = fir.load %[[VAL_5]]#0 : !fir.ref<i32>
+!CHECK:  %[[VAL_7:.*]] = arith.constant 1 : i32
+!CHECK:  %[[VAL_8:.*]] = arith.addi %[[VAL_6]], %[[VAL_7]] : i32
+!CHECK:  hlfir.assign %[[VAL_8]] to %[[VAL_5]]#0 : i32, !fir.ref<i32>
+!CHECK:  omp.terminator
+!CHECK:  }
+!CHECK:  omp.terminator
+!CHECK:  }
+!CHECK:  omp.terminator
+!CHECK:  }
+!CHECK:  return
+!CHECK:  }
+
+subroutine in_reduction
+   integer :: x
+   x = 0
+   !$omp parallel
+   !$omp taskgroup task_reduction(+:x)
+   !$omp task in_reduction(+:x)
+   x = x + 1
+   !$omp end task
+   !$omp end taskgroup
+   !$omp end parallel
+end subroutine
+
diff --git a/flang/test/Lower/OpenMP/task_reduction.f90 b/flang/test/Lower/OpenMP/task_reduction.f90
new file mode 100644
index 00000000000000..70f64ba34d3405
--- /dev/null
+++ b/flang/test/Lower/OpenMP/task_reduction.f90
@@ -0,0 +1,43 @@
+! 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.declare_reduction
+!CHECK-SAME: @[[RED_I32_NAME:.*]] : i32 init {
+!CHECK: ^bb0(%{{.*}}: i32):
+!CHECK:  %[[C0_1:.*]] = arith.constant 0 : i32
+!CHECK:  omp.yield(%[[C0_1]] : i32)
+!CHECK: } combiner {
+!CHECK: ^bb0(%[[ARG0:.*]]: i32, %[[ARG1:.*]]: i32):
+!CHECK:  %[[RES:.*]] = arith.addi %[[ARG0]], %[[ARG1]] : i32
+!CHECK:  omp.yield(%[[RES]] : i32)
+!CHECK: }
+
+!CHECK-LABEL: func.func @_QPtest_add() {
+!CHECK:  %[[VAL_0:.*]] = fir.address_of(@_QFtest_addEx) : !fir.ref<i32>
+!CHECK:  %[[VAL_1:.*]]:2 = hlfir.declare %[[VAL_0]] {uniq_name = "_QFtest_addEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+!CHECK:  omp.taskgroup task_reduction(@[[RED_I32_NAME]]  %[[VAL_1]]#0 -> %[[VAL_2:.*]] : !fir.ref<i32>) {
+!CHECK:  omp.task {
+!CHECK:  %[[VAL_3:.*]] = fir.alloca i32 {bindc_name = "x", pinned, uniq_name = "_QFtest_addEx"}
+!CHECK:  %[[VAL_4:.*]]:2 =  hlfir.declare %[[VAL_3]] {uniq_name = "_QFtest_addEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+!CHECK:  %[[VAL_5:.*]] = fir.load %[[VAL_1]]#0 : !fir.ref<i32>
+!CHECK:  hlfir.assign %[[VAL_5]] to %[[VAL_4]]#0 : i32, !fir.ref<i32>
+!CHECK:  %[[VAL_6:.*]] = fir.load %[[VAL_4]]#0 : !fir.ref<i32>
+!CHECK:  %[[VAL_7:.*]] = arith.constant 1 : i32
+!CHECK:  %[[VAL_8:.*]] = arith.addi %[[VAL_6]], %[[VAL_7]] : i32
+!CHECK:  hlfir.assign %[[VAL_8]] to %[[VAL_4]]#0 : i32, !fir.ref<i32>
+!CHECK:  omp.terminator
+!CHECK:  }
+!CHECK:  omp.terminator
+!CHECK:  }
+!CHECK:  return
+!CHECK:  }
+
+subroutine test_add
+   integer :: x = 0
+   !$omp taskgroup task_reduction(+:x)
+   !$omp task
+   x = x + 1
+   !$omp end task
+   !$omp end taskgroup
+end subroutine
+



More information about the flang-commits mailing list