[Mlir-commits] [flang] [llvm] [mlir] [flang][MLIR][OpenMP] make reduction by-ref toggled per variable (PR #92244)

Tom Eccles llvmlistbot at llvm.org
Wed May 15 03:52:30 PDT 2024


https://github.com/tblah created https://github.com/llvm/llvm-project/pull/92244

Fixes #88935

Toggling reduction by-ref broke when multiple reduction clauses were used. Decisions made for the by-ref status for later clauses could then invalidate decisions for earlier clauses. For example,

```
reduction(+:scalar,scalar2) reduction(+:array)
```

The first clause would choose by value reduction and generate by-value reduction regions, but then after this the second clause would force by-ref to support the array argument. But by the time the second clause is processed, the first clause has already had the wrong kind of reduction regions generated.

This is solved by toggling whether a variable should be reduced by reference per variable. In the above example, this allows only `array` to be reduced by ref.

>From 4880fd3051f7359d0ac38cb8441546f62dba0750 Mon Sep 17 00:00:00 2001
From: Tom Eccles <tom.eccles at arm.com>
Date: Fri, 3 May 2024 16:28:13 +0000
Subject: [PATCH] [flang][MLIR][OpenMP] make reduction by-ref toggled per
 variable

Fixes #88935

Toggling reduction by-ref broke when multiple reduction clauses were
used. Decisions made for the by-ref status for later clauses could then
invalidate decisions for earlier clauses. For example,

```
reduction(+:scalar,scalar2) reduction(+:array)
```

The first clause would choose by value reduction and generate by-value
reduction regions, but then after this the second clause would force
by-ref to support the array argument. But by the time the second clause
is processed, the first clause has already had the wrong kind of
reduction regions generated.

This is solved by toggling whether a variable should be reduced by
reference per variable. In the above example, this allows only `array`
to be reduced by ref.
---
 flang/lib/Lower/OpenMP/ClauseProcessor.cpp    |  12 +-
 flang/lib/Lower/OpenMP/OpenMP.cpp             |   6 -
 flang/lib/Lower/OpenMP/ReductionProcessor.cpp |  29 ++--
 flang/lib/Lower/OpenMP/ReductionProcessor.h   |   4 +-
 .../test/Fir/omp-reduction-embox-codegen.fir  |   2 +-
 .../Lower/OpenMP/default-clause-byref.f90     |   2 +-
 .../delayed-privatization-reduction-byref.f90 |   2 +-
 .../OpenMP/parallel-reduction-add-byref.f90   |   6 +-
 .../parallel-reduction-allocatable-array.f90  |   2 +-
 .../OpenMP/parallel-reduction-array-lb.f90    |   2 +-
 .../Lower/OpenMP/parallel-reduction-array.f90 |   2 +-
 .../OpenMP/parallel-reduction-array2.f90      |   2 +-
 .../Lower/OpenMP/parallel-reduction-byref.f90 |   2 +-
 .../test/Lower/OpenMP/parallel-reduction3.f90 |   2 +-
 .../parallel-wsloop-reduction-byref.f90       |   2 +-
 .../OpenMP/wsloop-reduction-add-byref.f90     |  14 +-
 .../wsloop-reduction-add-hlfir-byref.f90      |   2 +-
 .../OpenMP/wsloop-reduction-allocatable.f90   |   2 +-
 .../wsloop-reduction-array-assumed-shape.f90  |   2 +-
 .../Lower/OpenMP/wsloop-reduction-array.f90   |   2 +-
 .../Lower/OpenMP/wsloop-reduction-array2.f90  |   2 +-
 .../OpenMP/wsloop-reduction-iand-byref.f90    |   2 +-
 .../OpenMP/wsloop-reduction-ieor-byref.f90    |   2 +-
 .../OpenMP/wsloop-reduction-ior-byref.f90     |   2 +-
 .../wsloop-reduction-logical-and-byref.f90    |   6 +-
 .../wsloop-reduction-logical-eqv-byref.f90    |   6 +-
 .../wsloop-reduction-logical-neqv-byref.f90   |   6 +-
 .../wsloop-reduction-logical-or-byref.f90     |   6 +-
 .../OpenMP/wsloop-reduction-max-2-byref.f90   |   2 +-
 .../OpenMP/wsloop-reduction-max-byref.f90     |   6 +-
 .../wsloop-reduction-max-hlfir-byref.f90      |   2 +-
 .../OpenMP/wsloop-reduction-min-byref.f90     |   6 +-
 .../OpenMP/wsloop-reduction-mul-byref.f90     |  14 +-
 .../wsloop-reduction-multiple-clauses.f90     | 164 ++++++++++++++++++
 .../llvm/Frontend/OpenMP/OMPIRBuilder.h       |   2 +-
 llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp     |  18 +-
 .../Frontend/OpenMPIRBuilderTest.cpp          |  12 +-
 .../Dialect/OpenMP/OpenMPClauseOperands.h     |   2 +-
 mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td |  59 +++----
 .../Conversion/SCFToOpenMP/SCFToOpenMP.cpp    |   5 +
 mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp  | 120 +++++++++----
 .../OpenMP/OpenMPToLLVMIRTranslation.cpp      |  43 +++--
 mlir/test/Dialect/OpenMP/ops.mlir             |  38 ++++
 .../openmp-parallel-reduction-cleanup.mlir    |   2 +-
 .../Target/LLVMIR/openmp-reduction-byref.mlir |   2 +-
 .../LLVMIR/openmp-reduction-init-arg.mlir     |   2 +-
 .../openmp-wsloop-reduction-cleanup.mlir      |   2 +-
 47 files changed, 441 insertions(+), 191 deletions(-)
 create mode 100644 flang/test/Lower/OpenMP/wsloop-reduction-multiple-clauses.f90

diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index 0ea87314d571f..a57b96e365999 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -933,20 +933,18 @@ bool ClauseProcessor::processReduction(
   return findRepeatableClause<omp::clause::Reduction>(
       [&](const omp::clause::Reduction &clause,
           const Fortran::parser::CharBlock &) {
-        // Use local lists of reductions to prevent variables from other
-        // already-processed reduction clauses from impacting this reduction.
-        // For example, the whole `reductionVars` array is queried to decide
-        // whether to do the reduction byref.
         llvm::SmallVector<mlir::Value> reductionVars;
+        llvm::SmallVector<bool> reduceVarByRef;
         llvm::SmallVector<mlir::Attribute> reductionDeclSymbols;
         llvm::SmallVector<const Fortran::semantics::Symbol *> reductionSyms;
         ReductionProcessor rp;
-        rp.addDeclareReduction(currentLocation, converter, clause,
-                               reductionVars, reductionDeclSymbols,
-                               outReductionSyms ? &reductionSyms : nullptr);
+        rp.addDeclareReduction(
+            currentLocation, converter, clause, reductionVars, reduceVarByRef,
+            reductionDeclSymbols, outReductionSyms ? &reductionSyms : nullptr);
 
         // Copy local lists into the output.
         llvm::copy(reductionVars, std::back_inserter(result.reductionVars));
+        llvm::copy(reduceVarByRef, std::back_inserter(result.reduceVarByRef));
         llvm::copy(reductionDeclSymbols,
                    std::back_inserter(result.reductionDeclSymbols));
 
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index f21acdd64d7c3..aaf0c6501ab20 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -1007,8 +1007,6 @@ static void genParallelClauses(
 
   if (processReduction) {
     cp.processReduction(loc, clauseOps, &reductionTypes, &reductionSyms);
-    if (ReductionProcessor::doReductionByRef(clauseOps.reductionVars))
-      clauseOps.reductionByRefAttr = converter.getFirOpBuilder().getUnitAttr();
   }
 }
 
@@ -1200,7 +1198,6 @@ static void genWsloopClauses(
     mlir::Location loc, mlir::omp::WsloopClauseOps &clauseOps,
     llvm::SmallVectorImpl<mlir::Type> &reductionTypes,
     llvm::SmallVectorImpl<const Fortran::semantics::Symbol *> &reductionSyms) {
-  fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
   ClauseProcessor cp(converter, semaCtx, clauses);
   cp.processNowait(clauseOps);
   cp.processOrdered(clauseOps);
@@ -1208,9 +1205,6 @@ static void genWsloopClauses(
   cp.processSchedule(stmtCtx, clauseOps);
   // TODO Support delayed privatization.
 
-  if (ReductionProcessor::doReductionByRef(clauseOps.reductionVars))
-    clauseOps.reductionByRefAttr = firOpBuilder.getUnitAttr();
-
   cp.processTODO<clause::Allocate, clause::Linear, clause::Order>(
       loc, llvm::omp::Directive::OMPD_do);
 }
diff --git a/flang/lib/Lower/OpenMP/ReductionProcessor.cpp b/flang/lib/Lower/OpenMP/ReductionProcessor.cpp
index b3f08eb81c799..689f3adc0a429 100644
--- a/flang/lib/Lower/OpenMP/ReductionProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ReductionProcessor.cpp
@@ -657,25 +657,17 @@ mlir::omp::DeclareReductionOp ReductionProcessor::createDeclareReduction(
   return decl;
 }
 
-// TODO: By-ref vs by-val reductions are currently toggled for the whole
-//       operation (possibly effecting multiple reduction variables).
-//       This could cause a problem with openmp target reductions because
-//       by-ref trivial types may not be supported.
-bool ReductionProcessor::doReductionByRef(
-    const llvm::SmallVectorImpl<mlir::Value> &reductionVars) {
-  if (reductionVars.empty())
-    return false;
+static bool doReductionByRef(mlir::Value reductionVar) {
   if (forceByrefReduction)
     return true;
 
-  for (mlir::Value reductionVar : reductionVars) {
-    if (auto declare =
-            mlir::dyn_cast<hlfir::DeclareOp>(reductionVar.getDefiningOp()))
-      reductionVar = declare.getMemref();
+  if (auto declare =
+          mlir::dyn_cast<hlfir::DeclareOp>(reductionVar.getDefiningOp()))
+    reductionVar = declare.getMemref();
+
+  if (!fir::isa_trivial(fir::unwrapRefType(reductionVar.getType())))
+    return true;
 
-    if (!fir::isa_trivial(fir::unwrapRefType(reductionVar.getType())))
-      return true;
-  }
   return false;
 }
 
@@ -684,6 +676,7 @@ void ReductionProcessor::addDeclareReduction(
     Fortran::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 Fortran::semantics::Symbol *>
         *reductionSymbols) {
@@ -764,8 +757,8 @@ void ReductionProcessor::addDeclareReduction(
            "reduction input var is a reference");
 
     reductionVars.push_back(symVal);
+    reduceVarByRef.push_back(doReductionByRef(symVal));
   }
-  const bool isByRef = doReductionByRef(reductionVars);
 
   if (const auto &redDefinedOp =
           std::get_if<omp::clause::DefinedOperator>(&redOperator.u)) {
@@ -787,7 +780,7 @@ void ReductionProcessor::addDeclareReduction(
       break;
     }
 
-    for (mlir::Value symVal : reductionVars) {
+    for (auto [symVal, isByRef] : llvm::zip(reductionVars, reduceVarByRef)) {
       auto redType = mlir::cast<fir::ReferenceType>(symVal.getType());
       const auto &kindMap = firOpBuilder.getKindMap();
       if (mlir::isa<fir::LogicalType>(redType.getEleTy()))
@@ -811,7 +804,7 @@ void ReductionProcessor::addDeclareReduction(
             *reductionIntrinsic)) {
       ReductionProcessor::ReductionIdentifier redId =
           ReductionProcessor::getReductionType(*reductionIntrinsic);
-      for (mlir::Value symVal : reductionVars) {
+      for (auto [symVal, isByRef] : llvm::zip(reductionVars, reduceVarByRef)) {
         auto redType = mlir::cast<fir::ReferenceType>(symVal.getType());
         if (!redType.getEleTy().isIntOrIndexOrFloat())
           TODO(currentLocation,
diff --git a/flang/lib/Lower/OpenMP/ReductionProcessor.h b/flang/lib/Lower/OpenMP/ReductionProcessor.h
index 8b116a4c52041..95d77c8154415 100644
--- a/flang/lib/Lower/OpenMP/ReductionProcessor.h
+++ b/flang/lib/Lower/OpenMP/ReductionProcessor.h
@@ -73,9 +73,6 @@ class ReductionProcessor {
   static const Fortran::semantics::SourceName
   getRealName(const omp::clause::ProcedureDesignator &pd);
 
-  static bool
-  doReductionByRef(const llvm::SmallVectorImpl<mlir::Value> &reductionVars);
-
   static std::string getReductionName(llvm::StringRef name,
                                       const fir::KindMapping &kindMap,
                                       mlir::Type ty, bool isByRef);
@@ -128,6 +125,7 @@ class ReductionProcessor {
       Fortran::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 Fortran::semantics::Symbol *>
           *reductionSymbols = nullptr);
diff --git a/flang/test/Fir/omp-reduction-embox-codegen.fir b/flang/test/Fir/omp-reduction-embox-codegen.fir
index 7602012ebc5c9..1645e1a407ad4 100644
--- a/flang/test/Fir/omp-reduction-embox-codegen.fir
+++ b/flang/test/Fir/omp-reduction-embox-codegen.fir
@@ -25,7 +25,7 @@ omp.declare_reduction @test_reduction : !fir.ref<!fir.box<i32>> init {
 
 func.func @_QQmain() attributes {fir.bindc_name = "reduce"} {
   %4 = fir.alloca !fir.box<i32>
-  omp.parallel byref reduction(@test_reduction %4 -> %arg0 : !fir.ref<!fir.box<i32>>) {
+  omp.parallel reduction(byref @test_reduction %4 -> %arg0 : !fir.ref<!fir.box<i32>>) {
     omp.terminator
   }
   return
diff --git a/flang/test/Lower/OpenMP/default-clause-byref.f90 b/flang/test/Lower/OpenMP/default-clause-byref.f90
index 7cc2bc2e0c710..7893c4d7d5732 100644
--- a/flang/test/Lower/OpenMP/default-clause-byref.f90
+++ b/flang/test/Lower/OpenMP/default-clause-byref.f90
@@ -351,7 +351,7 @@ subroutine skipped_default_clause_checks()
        type(it)::iii
 
 !CHECK: omp.parallel {
-!CHECK: omp.wsloop byref reduction(@min_byref_i32 %[[VAL_Z_DECLARE]]#0 -> %[[PRV:.+]] : !fir.ref<i32>) {
+!CHECK: omp.wsloop reduction(byref @min_byref_i32 %[[VAL_Z_DECLARE]]#0 -> %[[PRV:.+]] : !fir.ref<i32>) {
 !CHECK-NEXT: omp.loop_nest (%[[ARG:.*]]) {{.*}} {
 !CHECK: omp.yield
 !CHECK: }
diff --git a/flang/test/Lower/OpenMP/delayed-privatization-reduction-byref.f90 b/flang/test/Lower/OpenMP/delayed-privatization-reduction-byref.f90
index 49d1142ea4b6a..72e91680a4310 100644
--- a/flang/test/Lower/OpenMP/delayed-privatization-reduction-byref.f90
+++ b/flang/test/Lower/OpenMP/delayed-privatization-reduction-byref.f90
@@ -26,5 +26,5 @@ subroutine red_and_delayed_private
 
 ! CHECK-LABEL: _QPred_and_delayed_private
 ! CHECK: omp.parallel
-! CHECK-SAME: reduction(@[[REDUCTION_SYM]] %{{.*}} -> %arg0 : !fir.ref<i32>)
+! CHECK-SAME: reduction(byref @[[REDUCTION_SYM]] %{{.*}} -> %arg0 : !fir.ref<i32>)
 ! CHECK-SAME: private(@[[PRIVATIZER_SYM]] %{{.*}} -> %arg1 : !fir.ref<i32>) {
diff --git a/flang/test/Lower/OpenMP/parallel-reduction-add-byref.f90 b/flang/test/Lower/OpenMP/parallel-reduction-add-byref.f90
index 2a1d26407b27e..7347d9324feac 100644
--- a/flang/test/Lower/OpenMP/parallel-reduction-add-byref.f90
+++ b/flang/test/Lower/OpenMP/parallel-reduction-add-byref.f90
@@ -40,7 +40,7 @@
 !CHECK:  %[[I_DECL:.*]]:2 = hlfir.declare %[[IREF]] {uniq_name = "_QFsimple_int_addEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 !CHECK:  %[[I_START:.*]] = arith.constant 0 : i32
 !CHECK:  hlfir.assign %[[I_START]] to %[[I_DECL]]#0 : i32, !fir.ref<i32>
-!CHECK:  omp.parallel byref reduction(@[[RED_I32_NAME]] %[[I_DECL]]#0 -> %[[PRV:.+]] : !fir.ref<i32>) {
+!CHECK:  omp.parallel reduction(byref @[[RED_I32_NAME]] %[[I_DECL]]#0 -> %[[PRV:.+]] : !fir.ref<i32>) {
 !CHECK:    %[[P_DECL:.+]]:2 = hlfir.declare %[[PRV]] {{.*}} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 !CHECK:    %[[LPRV:.+]] = fir.load %[[P_DECL]]#0 : !fir.ref<i32>
 !CHECK:    %[[I_INCR:.*]] = arith.constant 1 : i32
@@ -65,7 +65,7 @@ subroutine simple_int_add
 !CHECK:  %[[R_DECL:.*]]:2 = hlfir.declare %[[RREF]] {uniq_name = "_QFsimple_real_addEr"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 !CHECK:  %[[R_START:.*]] = arith.constant 0.000000e+00 : f32
 !CHECK:  hlfir.assign %[[R_START]] to %[[R_DECL]]#0 : f32, !fir.ref<f32>
-!CHECK:  omp.parallel byref reduction(@[[RED_F32_NAME]] %[[R_DECL]]#0 -> %[[PRV:.+]] : !fir.ref<f32>) {
+!CHECK:  omp.parallel reduction(byref @[[RED_F32_NAME]] %[[R_DECL]]#0 -> %[[PRV:.+]] : !fir.ref<f32>) {
 !CHECK:    %[[P_DECL:.+]]:2 = hlfir.declare %[[PRV]] {{.*}} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 !CHECK:    %[[LPRV:.+]] = fir.load %[[P_DECL]]#0 : !fir.ref<f32>
 !CHECK:    %[[R_INCR:.*]] = arith.constant 1.500000e+00 : f32
@@ -94,7 +94,7 @@ subroutine simple_real_add
 !CHECK:  hlfir.assign %[[R_START]] to %[[R_DECL]]#0 : f32, !fir.ref<f32>
 !CHECK:  %[[I_START:.*]] = arith.constant 0 : i32
 !CHECK:  hlfir.assign %[[I_START]] to %[[I_DECL]]#0 : i32, !fir.ref<i32>
-!CHECK:  omp.parallel byref reduction(@[[RED_I32_NAME]] %[[I_DECL]]#0 -> %[[IPRV:.+]] : !fir.ref<i32>, @[[RED_F32_NAME]] %[[R_DECL]]#0 -> %[[RPRV:.+]] : !fir.ref<f32>) {
+!CHECK:  omp.parallel reduction(byref @[[RED_I32_NAME]] %[[I_DECL]]#0 -> %[[IPRV:.+]] : !fir.ref<i32>, byref @[[RED_F32_NAME]] %[[R_DECL]]#0 -> %[[RPRV:.+]] : !fir.ref<f32>) {
 !CHECK:    %[[IP_DECL:.+]]:2 = hlfir.declare %[[IPRV]] {{.*}} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 !CHECK:    %[[RP_DECL:.+]]:2 = hlfir.declare %[[RPRV]] {{.*}} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 !CHECK:    %[[R_INCR:.*]] = arith.constant 1.500000e+00 : f32
diff --git a/flang/test/Lower/OpenMP/parallel-reduction-allocatable-array.f90 b/flang/test/Lower/OpenMP/parallel-reduction-allocatable-array.f90
index 28216ef91c3a3..f6d3b0b73f738 100644
--- a/flang/test/Lower/OpenMP/parallel-reduction-allocatable-array.f90
+++ b/flang/test/Lower/OpenMP/parallel-reduction-allocatable-array.f90
@@ -95,7 +95,7 @@ program reduce
 ! CHECK:             %[[VAL_14:.*]] = arith.constant 0 : i32
 ! CHECK:             %[[VAL_15:.*]] = arith.constant 10 : i32
 ! CHECK:             %[[VAL_16:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_box_heap_Uxi32 %[[VAL_3]]#0 -> %[[VAL_17:.*]] : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_box_heap_Uxi32 %[[VAL_3]]#0 -> %[[VAL_17:.*]] : !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_18:.*]]) : i32 = (%[[VAL_14]]) to (%[[VAL_15]]) inclusive step (%[[VAL_16]]) {
 ! CHECK:                 %[[VAL_19:.*]]:2 = hlfir.declare %[[VAL_17]] {fortran_attrs = {{.*}}<allocatable>, uniq_name = "_QFEr"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xi32>>>>)
 ! CHECK:                 fir.store %[[VAL_18]] to %[[VAL_13]]#1 : !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/parallel-reduction-array-lb.f90 b/flang/test/Lower/OpenMP/parallel-reduction-array-lb.f90
index 8202e6d897157..b44fe4c1f4cc2 100644
--- a/flang/test/Lower/OpenMP/parallel-reduction-array-lb.f90
+++ b/flang/test/Lower/OpenMP/parallel-reduction-array-lb.f90
@@ -77,7 +77,7 @@ program reduce
 ! CHECK:           %[[VAL_6:.*]]:2 = hlfir.declare %[[VAL_0]](%[[VAL_5]]) {uniq_name = "_QFEi"} : (!fir.ref<!fir.array<3x2xi32>>, !fir.shapeshift<2>) -> (!fir.box<!fir.array<3x2xi32>>, !fir.ref<!fir.array<3x2xi32>>)
 ! CHECK:           %[[VAL_7:.*]] = fir.alloca !fir.box<!fir.array<3x2xi32>>
 ! CHECK:           fir.store %[[VAL_6]]#0 to %[[VAL_7]] : !fir.ref<!fir.box<!fir.array<3x2xi32>>>
-! CHECK:           omp.parallel byref reduction(@add_reduction_byref_box_3x2xi32 %[[VAL_7]] -> %[[VAL_8:.*]] : !fir.ref<!fir.box<!fir.array<3x2xi32>>>) {
+! CHECK:           omp.parallel reduction(byref @add_reduction_byref_box_3x2xi32 %[[VAL_7]] -> %[[VAL_8:.*]] : !fir.ref<!fir.box<!fir.array<3x2xi32>>>) {
 ! CHECK:             %[[VAL_9:.*]]:2 = hlfir.declare %[[VAL_8]] {uniq_name = "_QFEi"} : (!fir.ref<!fir.box<!fir.array<3x2xi32>>>) -> (!fir.ref<!fir.box<!fir.array<3x2xi32>>>, !fir.ref<!fir.box<!fir.array<3x2xi32>>>)
 ! CHECK:             %[[VAL_10:.*]] = arith.constant 3 : i32
 ! CHECK:             %[[VAL_11:.*]] = fir.load %[[VAL_9]]#0 : !fir.ref<!fir.box<!fir.array<3x2xi32>>>
diff --git a/flang/test/Lower/OpenMP/parallel-reduction-array.f90 b/flang/test/Lower/OpenMP/parallel-reduction-array.f90
index 34f4ee0a9eb3a..60b21c9b1ebbe 100644
--- a/flang/test/Lower/OpenMP/parallel-reduction-array.f90
+++ b/flang/test/Lower/OpenMP/parallel-reduction-array.f90
@@ -70,7 +70,7 @@ program reduce
 ! CHECK:           %[[VAL_4:.*]] = fir.embox %[[VAL_3]]#0(%[[VAL_2]]) : (!fir.ref<!fir.array<3xi32>>, !fir.shape<1>) -> !fir.box<!fir.array<3xi32>>
 ! CHECK:           %[[VAL_5:.*]] = fir.alloca !fir.box<!fir.array<3xi32>>
 ! CHECK:           fir.store %[[VAL_4]] to %[[VAL_5]] : !fir.ref<!fir.box<!fir.array<3xi32>>>
-! CHECK:           omp.parallel byref reduction(@add_reduction_byref_box_3xi32 %[[VAL_5]] -> %[[VAL_6:.*]] : !fir.ref<!fir.box<!fir.array<3xi32>>>) {
+! CHECK:           omp.parallel reduction(byref @add_reduction_byref_box_3xi32 %[[VAL_5]] -> %[[VAL_6:.*]] : !fir.ref<!fir.box<!fir.array<3xi32>>>) {
 ! CHECK:             %[[VAL_7:.*]]:2 = hlfir.declare %[[VAL_6]] {uniq_name = "_QFEi"} : (!fir.ref<!fir.box<!fir.array<3xi32>>>) -> (!fir.ref<!fir.box<!fir.array<3xi32>>>, !fir.ref<!fir.box<!fir.array<3xi32>>>)
 ! CHECK:             %[[VAL_8:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_9:.*]] = fir.load %[[VAL_7]]#0 : !fir.ref<!fir.box<!fir.array<3xi32>>>
diff --git a/flang/test/Lower/OpenMP/parallel-reduction-array2.f90 b/flang/test/Lower/OpenMP/parallel-reduction-array2.f90
index aa14092554eda..5d4c86d1d76e8 100644
--- a/flang/test/Lower/OpenMP/parallel-reduction-array2.f90
+++ b/flang/test/Lower/OpenMP/parallel-reduction-array2.f90
@@ -69,7 +69,7 @@ program reduce
 ! CHECK:           %[[VAL_4:.*]] = fir.embox %[[VAL_3]]#0(%[[VAL_2]]) : (!fir.ref<!fir.array<3xi32>>, !fir.shape<1>) -> !fir.box<!fir.array<3xi32>>
 ! CHECK:           %[[VAL_5:.*]] = fir.alloca !fir.box<!fir.array<3xi32>>
 ! CHECK:           fir.store %[[VAL_4]] to %[[VAL_5]] : !fir.ref<!fir.box<!fir.array<3xi32>>>
-! CHECK:           omp.parallel byref reduction(@add_reduction_byref_box_3xi32 %[[VAL_5]] -> %[[VAL_6:.*]] : !fir.ref<!fir.box<!fir.array<3xi32>>>) {
+! CHECK:           omp.parallel reduction(byref @add_reduction_byref_box_3xi32 %[[VAL_5]] -> %[[VAL_6:.*]] : !fir.ref<!fir.box<!fir.array<3xi32>>>) {
 ! CHECK:             %[[VAL_7:.*]]:2 = hlfir.declare %[[VAL_6]] {uniq_name = "_QFEi"} : (!fir.ref<!fir.box<!fir.array<3xi32>>>) -> (!fir.ref<!fir.box<!fir.array<3xi32>>>, !fir.ref<!fir.box<!fir.array<3xi32>>>)
 ! CHECK:             %[[VAL_8:.*]] = fir.load %[[VAL_7]]#0 : !fir.ref<!fir.box<!fir.array<3xi32>>>
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : index
diff --git a/flang/test/Lower/OpenMP/parallel-reduction-byref.f90 b/flang/test/Lower/OpenMP/parallel-reduction-byref.f90
index fdcdf0c0b8d95..5685e2c584ace 100644
--- a/flang/test/Lower/OpenMP/parallel-reduction-byref.f90
+++ b/flang/test/Lower/OpenMP/parallel-reduction-byref.f90
@@ -21,7 +21,7 @@
 !CHECK:    %[[RED_ACCUM_DECL:[_a-z0-9]+]]:2 = hlfir.declare %[[RED_ACCUM_REF]] {uniq_name = "_QFEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 !CHECK:    %[[C0:[_a-z0-9]+]] = arith.constant 0 : i32
 !CHECK:    hlfir.assign %[[C0]] to %[[RED_ACCUM_DECL]]#0 : i32, !fir.ref<i32>
-!CHECK:    omp.parallel byref reduction(@[[REDUCTION_DECLARE]] %[[RED_ACCUM_DECL]]#0 -> %[[PRIVATE_RED:[a-z0-9]+]] : !fir.ref<i32>) {
+!CHECK:    omp.parallel reduction(byref @[[REDUCTION_DECLARE]] %[[RED_ACCUM_DECL]]#0 -> %[[PRIVATE_RED:[a-z0-9]+]] : !fir.ref<i32>) {
 !CHECK:      %[[PRIVATE_DECL:[_a-z0-9]+]]:2 = hlfir.declare %[[PRIVATE_RED]] {uniq_name = "_QFEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 !CHECK:      %[[C1:[_a-z0-9]+]] = arith.constant 1 : i32
 !CHECK:      hlfir.assign %[[C1]] to %[[PRIVATE_DECL]]#0 : i32, !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/parallel-reduction3.f90 b/flang/test/Lower/OpenMP/parallel-reduction3.f90
index 17d805c0d142b..47b743a558b49 100644
--- a/flang/test/Lower/OpenMP/parallel-reduction3.f90
+++ b/flang/test/Lower/OpenMP/parallel-reduction3.f90
@@ -74,7 +74,7 @@
 ! CHECK:             %[[VAL_18:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_19:.*]] = fir.alloca !fir.box<!fir.array<?xi32>>
 ! CHECK:             fir.store %[[VAL_12]]#0 to %[[VAL_19]] : !fir.ref<!fir.box<!fir.array<?xi32>>>
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_box_Uxi32 %[[VAL_19]] -> %[[VAL_20:.*]] : !fir.ref<!fir.box<!fir.array<?xi32>>>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_box_Uxi32 %[[VAL_19]] -> %[[VAL_20:.*]] : !fir.ref<!fir.box<!fir.array<?xi32>>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_21:.*]]) : i32 = (%[[VAL_16]]) to (%[[VAL_17]]) inclusive step (%[[VAL_18]]) {
 ! CHECK:                 %[[VAL_22:.*]]:2 = hlfir.declare %[[VAL_20]] {uniq_name = "_QFsEc"} : (!fir.ref<!fir.box<!fir.array<?xi32>>>) -> (!fir.ref<!fir.box<!fir.array<?xi32>>>, !fir.ref<!fir.box<!fir.array<?xi32>>>)
 ! CHECK:                 fir.store %[[VAL_21]] to %[[VAL_15]]#1 : !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/parallel-wsloop-reduction-byref.f90 b/flang/test/Lower/OpenMP/parallel-wsloop-reduction-byref.f90
index 66c80c31917ba..32caac39778de 100644
--- a/flang/test/Lower/OpenMP/parallel-wsloop-reduction-byref.f90
+++ b/flang/test/Lower/OpenMP/parallel-wsloop-reduction-byref.f90
@@ -4,7 +4,7 @@
 ! RUN: flang-new -fc1 -fopenmp -mmlir --force-byref-reduction -emit-hlfir %s -o - | FileCheck %s
 
 ! CHECK: omp.parallel {
-! CHECK: omp.wsloop byref reduction(@add_reduction_byref_i32
+! CHECK: omp.wsloop reduction(byref @add_reduction_byref_i32
 subroutine sb
   integer :: x
   x = 0
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-add-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-add-byref.f90
index c9d03435d9e18..cc4d055b3ea1f 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-add-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-add-byref.f90
@@ -82,7 +82,7 @@
 ! CHECK:             %[[VAL_7:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_8:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<i32>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<i32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_11:.*]]) : i32 = (%[[VAL_7]]) to (%[[VAL_8]]) inclusive step (%[[VAL_9]]) {
 ! CHECK:                 %[[VAL_12:.*]]:2 = hlfir.declare %[[VAL_10]] {uniq_name = "_QFsimple_int_reductionEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 fir.store %[[VAL_11]] to %[[VAL_6]]#1 : !fir.ref<i32>
@@ -125,7 +125,7 @@ subroutine simple_int_reduction
 ! CHECK:             %[[VAL_7:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_8:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_f32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<f32>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_f32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<f32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_11:.*]]) : i32 = (%[[VAL_7]]) to (%[[VAL_8]]) inclusive step (%[[VAL_9]]) {
 ! CHECK:                 %[[VAL_12:.*]]:2 = hlfir.declare %[[VAL_10]] {uniq_name = "_QFsimple_real_reductionEx"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 ! CHECK:                 fir.store %[[VAL_11]] to %[[VAL_6]]#1 : !fir.ref<i32>
@@ -169,7 +169,7 @@ subroutine simple_real_reduction
 ! CHECK:             %[[VAL_7:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_8:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<i32>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<i32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_11:.*]]) : i32 = (%[[VAL_7]]) to (%[[VAL_8]]) inclusive step (%[[VAL_9]]) {
 ! CHECK:                 %[[VAL_12:.*]]:2 = hlfir.declare %[[VAL_10]] {uniq_name = "_QFsimple_int_reduction_switch_orderEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 fir.store %[[VAL_11]] to %[[VAL_6]]#1 : !fir.ref<i32>
@@ -211,7 +211,7 @@ subroutine simple_int_reduction_switch_order
 ! CHECK:             %[[VAL_7:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_8:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_f32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<f32>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_f32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<f32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_11:.*]]) : i32 = (%[[VAL_7]]) to (%[[VAL_8]]) inclusive step (%[[VAL_9]]) {
 ! CHECK:                 %[[VAL_12:.*]]:2 = hlfir.declare %[[VAL_10]] {uniq_name = "_QFsimple_real_reduction_switch_orderEx"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 ! CHECK:                 fir.store %[[VAL_11]] to %[[VAL_6]]#1 : !fir.ref<i32>
@@ -262,7 +262,7 @@ subroutine simple_real_reduction_switch_order
 ! CHECK:             %[[VAL_13:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_14:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_15:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_16:.*]] : !fir.ref<i32>, @add_reduction_byref_i32 %[[VAL_5]]#0 -> %[[VAL_17:.*]] : !fir.ref<i32>, @add_reduction_byref_i32 %[[VAL_7]]#0 -> %[[VAL_18:.*]] : !fir.ref<i32>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_16:.*]] : !fir.ref<i32>, byref @add_reduction_byref_i32 %[[VAL_5]]#0 -> %[[VAL_17:.*]] : !fir.ref<i32>, byref @add_reduction_byref_i32 %[[VAL_7]]#0 -> %[[VAL_18:.*]] : !fir.ref<i32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_19:.*]]) : i32 = (%[[VAL_13]]) to (%[[VAL_14]]) inclusive step (%[[VAL_15]]) {
 ! CHECK:                 %[[VAL_20:.*]]:2 = hlfir.declare %[[VAL_16]] {uniq_name = "_QFmultiple_int_reductions_same_typeEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 %[[VAL_21:.*]]:2 = hlfir.declare %[[VAL_17]] {uniq_name = "_QFmultiple_int_reductions_same_typeEy"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
@@ -326,7 +326,7 @@ subroutine multiple_int_reductions_same_type
 ! CHECK:             %[[VAL_13:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_14:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_15:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_f32 %[[VAL_3]]#0 -> %[[VAL_16:.*]] : !fir.ref<f32>, @add_reduction_byref_f32 %[[VAL_5]]#0 -> %[[VAL_17:.*]] : !fir.ref<f32>, @add_reduction_byref_f32 %[[VAL_7]]#0 -> %[[VAL_18:.*]] : !fir.ref<f32>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_f32 %[[VAL_3]]#0 -> %[[VAL_16:.*]] : !fir.ref<f32>, byref @add_reduction_byref_f32 %[[VAL_5]]#0 -> %[[VAL_17:.*]] : !fir.ref<f32>, byref @add_reduction_byref_f32 %[[VAL_7]]#0 -> %[[VAL_18:.*]] : !fir.ref<f32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_19:.*]]) : i32 = (%[[VAL_13]]) to (%[[VAL_14]]) inclusive step (%[[VAL_15]]) {
 ! CHECK:                 %[[VAL_20:.*]]:2 = hlfir.declare %[[VAL_16]] {uniq_name = "_QFmultiple_real_reductions_same_typeEx"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 ! CHECK:                 %[[VAL_21:.*]]:2 = hlfir.declare %[[VAL_17]] {uniq_name = "_QFmultiple_real_reductions_same_typeEy"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
@@ -397,7 +397,7 @@ subroutine multiple_real_reductions_same_type
 ! CHECK:             %[[VAL_16:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_17:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_18:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_i32 %[[VAL_5]]#0 -> %[[VAL_19:.*]] : !fir.ref<i32>, @add_reduction_byref_i64 %[[VAL_7]]#0 -> %[[VAL_20:.*]] : !fir.ref<i64>, @add_reduction_byref_f32 %[[VAL_9]]#0 -> %[[VAL_21:.*]] : !fir.ref<f32>, @add_reduction_byref_f64 %[[VAL_3]]#0 -> %[[VAL_22:.*]] : !fir.ref<f64>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_i32 %[[VAL_5]]#0 -> %[[VAL_19:.*]] : !fir.ref<i32>, byref @add_reduction_byref_i64 %[[VAL_7]]#0 -> %[[VAL_20:.*]] : !fir.ref<i64>, byref @add_reduction_byref_f32 %[[VAL_9]]#0 -> %[[VAL_21:.*]] : !fir.ref<f32>, byref @add_reduction_byref_f64 %[[VAL_3]]#0 -> %[[VAL_22:.*]] : !fir.ref<f64>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_23:.*]]) : i32 = (%[[VAL_16]]) to (%[[VAL_17]]) inclusive step (%[[VAL_18]]) {
 ! CHECK:                 %[[VAL_24:.*]]:2 = hlfir.declare %[[VAL_19]] {uniq_name = "_QFmultiple_reductions_different_typeEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 %[[VAL_25:.*]]:2 = hlfir.declare %[[VAL_20]] {uniq_name = "_QFmultiple_reductions_different_typeEy"} : (!fir.ref<i64>) -> (!fir.ref<i64>, !fir.ref<i64>)
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-add-hlfir-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-add-hlfir-byref.f90
index 6a09fece80ae9..ec9abba41db0e 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-add-hlfir-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-add-hlfir-byref.f90
@@ -31,7 +31,7 @@
 ! CHECK:             %[[VAL_7:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_8:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<i32>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<i32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_11:.*]]) : i32 = (%[[VAL_7]]) to (%[[VAL_8]]) inclusive step (%[[VAL_9]]) {
 ! CHECK:                 %[[VAL_12:.*]]:2 = hlfir.declare %[[VAL_10]] {uniq_name = "_QFsimple_int_reductionEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 fir.store %[[VAL_11]] to %[[VAL_6]]#1 : !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-allocatable.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-allocatable.f90
index 8310aeb23df08..ccf186cf9908d 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-allocatable.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-allocatable.f90
@@ -80,7 +80,7 @@ program reduce
 ! CHECK:             %[[VAL_11:.*]] = arith.constant 0 : i32
 ! CHECK:             %[[VAL_12:.*]] = arith.constant 10 : i32
 ! CHECK:             %[[VAL_13:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_box_heap_i32 %[[VAL_5]]#0 -> %[[VAL_14:.*]] : !fir.ref<!fir.box<!fir.heap<i32>>>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_box_heap_i32 %[[VAL_5]]#0 -> %[[VAL_14:.*]] : !fir.ref<!fir.box<!fir.heap<i32>>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_15:.*]]) : i32 = (%[[VAL_11]]) to (%[[VAL_12]]) inclusive step (%[[VAL_13]]) {
 ! CHECK:                 %[[VAL_16:.*]]:2 = hlfir.declare %[[VAL_14]] {fortran_attrs = {{.*}}<allocatable>, uniq_name = "_QFEr"} : (!fir.ref<!fir.box<!fir.heap<i32>>>) -> (!fir.ref<!fir.box<!fir.heap<i32>>>, !fir.ref<!fir.box<!fir.heap<i32>>>)
 ! CHECK:                 fir.store %[[VAL_15]] to %[[VAL_10]]#1 : !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-array-assumed-shape.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-array-assumed-shape.f90
index 197800486c39e..be1e3ec0cb9d9 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-array-assumed-shape.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-array-assumed-shape.f90
@@ -84,7 +84,7 @@ subroutine reduce(r)
 ! CHECK:             %[[VAL_8:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_9:.*]] = fir.alloca !fir.box<!fir.array<?xf64>>
 ! CHECK:             fir.store %[[VAL_3]]#1 to %[[VAL_9]] : !fir.ref<!fir.box<!fir.array<?xf64>>>
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_box_Uxf64 %[[VAL_9]] -> %[[VAL_10:.*]] : !fir.ref<!fir.box<!fir.array<?xf64>>>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_box_Uxf64 %[[VAL_9]] -> %[[VAL_10:.*]] : !fir.ref<!fir.box<!fir.array<?xf64>>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_11:.*]]) : i32 = (%[[VAL_6]]) to (%[[VAL_7]]) inclusive step (%[[VAL_8]]) {
 ! CHECK:                 %[[VAL_12:.*]]:2 = hlfir.declare %[[VAL_10]] {fortran_attrs = {{.*}}, uniq_name = "_QFFreduceEr"} : (!fir.ref<!fir.box<!fir.array<?xf64>>>) -> (!fir.ref<!fir.box<!fir.array<?xf64>>>, !fir.ref<!fir.box<!fir.array<?xf64>>>)
 ! CHECK:                 fir.store %[[VAL_11]] to %[[VAL_5]]#1 : !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-array.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-array.f90
index 7aecf870cf800..add422f5d9564 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-array.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-array.f90
@@ -79,7 +79,7 @@ program reduce
 ! CHECK:             %[[VAL_11:.*]] = fir.embox %[[VAL_5]]#0(%[[VAL_4]]) : (!fir.ref<!fir.array<2xi32>>, !fir.shape<1>) -> !fir.box<!fir.array<2xi32>>
 ! CHECK:             %[[VAL_12:.*]] = fir.alloca !fir.box<!fir.array<2xi32>>
 ! CHECK:             fir.store %[[VAL_11]] to %[[VAL_12]] : !fir.ref<!fir.box<!fir.array<2xi32>>>
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_box_2xi32 %[[VAL_12]] -> %[[VAL_13:.*]] : !fir.ref<!fir.box<!fir.array<2xi32>>>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_box_2xi32 %[[VAL_12]] -> %[[VAL_13:.*]] : !fir.ref<!fir.box<!fir.array<2xi32>>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_14:.*]]) : i32 = (%[[VAL_8]]) to (%[[VAL_9]]) inclusive step (%[[VAL_10]]) {
 ! CHECK:                 %[[VAL_15:.*]]:2 = hlfir.declare %[[VAL_13]] {uniq_name = "_QFEr"} : (!fir.ref<!fir.box<!fir.array<2xi32>>>) -> (!fir.ref<!fir.box<!fir.array<2xi32>>>, !fir.ref<!fir.box<!fir.array<2xi32>>>)
 ! CHECK:                 fir.store %[[VAL_14]] to %[[VAL_7]]#1 : !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-array2.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-array2.f90
index d1d8a2c599444..e988567fc3371 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-array2.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-array2.f90
@@ -79,7 +79,7 @@ program reduce
 ! CHECK:             %[[VAL_11:.*]] = fir.embox %[[VAL_5]]#0(%[[VAL_4]]) : (!fir.ref<!fir.array<2xi32>>, !fir.shape<1>) -> !fir.box<!fir.array<2xi32>>
 ! CHECK:             %[[VAL_12:.*]] = fir.alloca !fir.box<!fir.array<2xi32>>
 ! CHECK:             fir.store %[[VAL_11]] to %[[VAL_12]] : !fir.ref<!fir.box<!fir.array<2xi32>>>
-! CHECK:             omp.wsloop byref reduction(@add_reduction_byref_box_2xi32 %[[VAL_12]] -> %[[VAL_13:.*]] : !fir.ref<!fir.box<!fir.array<2xi32>>>) {
+! CHECK:             omp.wsloop reduction(byref @add_reduction_byref_box_2xi32 %[[VAL_12]] -> %[[VAL_13:.*]] : !fir.ref<!fir.box<!fir.array<2xi32>>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_14:.*]]) : i32 = (%[[VAL_8]]) to (%[[VAL_9]]) inclusive step (%[[VAL_10]]) {
 ! CHECK:                 %[[VAL_15:.*]]:2 = hlfir.declare %[[VAL_13]] {uniq_name = "_QFEr"} : (!fir.ref<!fir.box<!fir.array<2xi32>>>) -> (!fir.ref<!fir.box<!fir.array<2xi32>>>, !fir.ref<!fir.box<!fir.array<2xi32>>>)
 ! CHECK:                 fir.store %[[VAL_14]] to %[[VAL_7]]#1 : !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-iand-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-iand-byref.f90
index df07a90653315..bcc71415f1dd1 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-iand-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-iand-byref.f90
@@ -35,7 +35,7 @@
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_10:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_11:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@iand_byref_i32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<i32>) {
+! CHECK:             omp.wsloop reduction(byref @iand_byref_i32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<i32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_13:.*]]) : i32 = (%[[VAL_9]]) to (%[[VAL_10]]) inclusive step (%[[VAL_11]]) {
 ! CHECK:                 %[[VAL_14:.*]]:2 = hlfir.declare %[[VAL_12]] {uniq_name = "_QFreduction_iandEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 fir.store %[[VAL_13]] to %[[VAL_8]]#1 : !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-ieor-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-ieor-byref.f90
index 50cec61b602bc..6825800b2cd87 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-ieor-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-ieor-byref.f90
@@ -28,7 +28,7 @@
 !CHECK: omp.parallel
 !CHECK: %[[I_REF:.*]] = fir.alloca i32 {adapt.valuebyref, pinned}
 !CHECK: %[[I_DECL:.*]]:2 = hlfir.declare %[[I_REF]] {uniq_name = "_QFreduction_ieorEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
-!CHECK: omp.wsloop byref reduction(@ieor_byref_i32 %[[X_DECL]]#0 -> %[[PRV:.+]] : !fir.ref<i32>)
+!CHECK: omp.wsloop reduction(byref @ieor_byref_i32 %[[X_DECL]]#0 -> %[[PRV:.+]] : !fir.ref<i32>)
 !CHECK-NEXT: omp.loop_nest
 !CHECK: %[[PRV_DECL:.+]]:2 = hlfir.declare %[[PRV]] {{.*}} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 !CHECK: fir.store %{{.*}} to %[[I_DECL]]#1 : !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-ior-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-ior-byref.f90
index d847bba89782d..00f4552917aa3 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-ior-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-ior-byref.f90
@@ -33,7 +33,7 @@
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_10:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_11:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@ior_byref_i32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<i32>) {
+! CHECK:             omp.wsloop reduction(byref @ior_byref_i32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<i32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_13:.*]]) : i32 = (%[[VAL_9]]) to (%[[VAL_10]]) inclusive step (%[[VAL_11]]) {
 ! CHECK:                 %[[VAL_14:.*]]:2 = hlfir.declare %[[VAL_12]] {uniq_name = "_QFreduction_iorEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 fir.store %[[VAL_13]] to %[[VAL_8]]#1 : !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-logical-and-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-logical-and-byref.f90
index 69789e4c751e0..331232a08e624 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-logical-and-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-logical-and-byref.f90
@@ -42,7 +42,7 @@
 ! CHECK:             %[[VAL_12:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_13:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_14:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@and_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
+! CHECK:             omp.wsloop reduction(byref @and_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_16:.*]]) : i32 = (%[[VAL_12]]) to (%[[VAL_13]]) inclusive step (%[[VAL_14]]) {
 ! CHECK:                 %[[VAL_17:.*]]:2 = hlfir.declare %[[VAL_15]] {uniq_name = "_QFsimple_reductionEx"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
 ! CHECK:                 fir.store %[[VAL_16]] to %[[VAL_11]]#1 : !fir.ref<i32>
@@ -92,7 +92,7 @@ end subroutine simple_reduction
 ! CHECK:             %[[VAL_12:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_13:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_14:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@and_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
+! CHECK:             omp.wsloop reduction(byref @and_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_16:.*]]) : i32 = (%[[VAL_12]]) to (%[[VAL_13]]) inclusive step (%[[VAL_14]]) {
 ! CHECK:                 %[[VAL_17:.*]]:2 = hlfir.declare %[[VAL_15]] {uniq_name = "_QFsimple_reduction_switch_orderEx"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
 ! CHECK:                 fir.store %[[VAL_16]] to %[[VAL_11]]#1 : !fir.ref<i32>
@@ -151,7 +151,7 @@ subroutine simple_reduction_switch_order(y)
 ! CHECK:             %[[VAL_20:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_21:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_22:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@and_reduction %[[VAL_7]]#0 -> %[[VAL_23:.*]] : !fir.ref<!fir.logical<4>>, @and_reduction %[[VAL_9]]#0 -> %[[VAL_24:.*]] : !fir.ref<!fir.logical<4>>, @and_reduction %[[VAL_11]]#0 -> %[[VAL_25:.*]] : !fir.ref<!fir.logical<4>>) {
+! CHECK:             omp.wsloop reduction(byref @and_reduction %[[VAL_7]]#0 -> %[[VAL_23:.*]] : !fir.ref<!fir.logical<4>>, byref @and_reduction %[[VAL_9]]#0 -> %[[VAL_24:.*]] : !fir.ref<!fir.logical<4>>, byref @and_reduction %[[VAL_11]]#0 -> %[[VAL_25:.*]] : !fir.ref<!fir.logical<4>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_26:.*]]) : i32 = (%[[VAL_20]]) to (%[[VAL_21]]) inclusive step (%[[VAL_22]]) {
 ! CHECK:                 %[[VAL_27:.*]]:2 = hlfir.declare %[[VAL_23]] {uniq_name = "_QFmultiple_reductionsEx"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
 ! CHECK:                 %[[VAL_28:.*]]:2 = hlfir.declare %[[VAL_24]] {uniq_name = "_QFmultiple_reductionsEy"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-logical-eqv-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-logical-eqv-byref.f90
index 54175994ecd8a..d1cb7b05f0af5 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-logical-eqv-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-logical-eqv-byref.f90
@@ -42,7 +42,7 @@
 ! CHECK:             %[[VAL_12:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_13:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_14:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@eqv_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
+! CHECK:             omp.wsloop reduction(byref @eqv_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_16:.*]]) : i32 = (%[[VAL_12]]) to (%[[VAL_13]]) inclusive step (%[[VAL_14]]) {
 ! CHECK:                 %[[VAL_17:.*]]:2 = hlfir.declare %[[VAL_15]] {uniq_name = "_QFsimple_reductionEx"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
 ! CHECK:                 fir.store %[[VAL_16]] to %[[VAL_11]]#1 : !fir.ref<i32>
@@ -91,7 +91,7 @@ subroutine simple_reduction(y)
 ! CHECK:             %[[VAL_12:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_13:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_14:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@eqv_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
+! CHECK:             omp.wsloop reduction(byref @eqv_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_16:.*]]) : i32 = (%[[VAL_12]]) to (%[[VAL_13]]) inclusive step (%[[VAL_14]]) {
 ! CHECK:                 %[[VAL_17:.*]]:2 = hlfir.declare %[[VAL_15]] {uniq_name = "_QFsimple_reduction_switch_orderEx"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
 ! CHECK:                 fir.store %[[VAL_16]] to %[[VAL_11]]#1 : !fir.ref<i32>
@@ -150,7 +150,7 @@ subroutine simple_reduction_switch_order(y)
 ! CHECK:             %[[VAL_20:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_21:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_22:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@eqv_reduction %[[VAL_7]]#0 -> %[[VAL_23:.*]] : !fir.ref<!fir.logical<4>>, @eqv_reduction %[[VAL_9]]#0 -> %[[VAL_24:.*]] : !fir.ref<!fir.logical<4>>, @eqv_reduction %[[VAL_11]]#0 -> %[[VAL_25:.*]] : !fir.ref<!fir.logical<4>>) {
+! CHECK:             omp.wsloop reduction(byref @eqv_reduction %[[VAL_7]]#0 -> %[[VAL_23:.*]] : !fir.ref<!fir.logical<4>>, byref @eqv_reduction %[[VAL_9]]#0 -> %[[VAL_24:.*]] : !fir.ref<!fir.logical<4>>, byref @eqv_reduction %[[VAL_11]]#0 -> %[[VAL_25:.*]] : !fir.ref<!fir.logical<4>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_26:.*]]) : i32 = (%[[VAL_20]]) to (%[[VAL_21]]) inclusive step (%[[VAL_22]]) {
 ! CHECK:                 %[[VAL_27:.*]]:2 = hlfir.declare %[[VAL_23]] {uniq_name = "_QFmultiple_reductionsEx"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
 ! CHECK:                 %[[VAL_28:.*]]:2 = hlfir.declare %[[VAL_24]] {uniq_name = "_QFmultiple_reductionsEy"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-logical-neqv-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-logical-neqv-byref.f90
index c0a82476c7b18..64c056c9c6bd2 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-logical-neqv-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-logical-neqv-byref.f90
@@ -42,7 +42,7 @@
 ! CHECK:             %[[VAL_12:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_13:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_14:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@neqv_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
+! CHECK:             omp.wsloop reduction(byref @neqv_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_16:.*]]) : i32 = (%[[VAL_12]]) to (%[[VAL_13]]) inclusive step (%[[VAL_14]]) {
 ! CHECK:                 %[[VAL_17:.*]]:2 = hlfir.declare %[[VAL_15]] {uniq_name = "_QFsimple_reductionEx"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
 ! CHECK:                 fir.store %[[VAL_16]] to %[[VAL_11]]#1 : !fir.ref<i32>
@@ -92,7 +92,7 @@ subroutine simple_reduction(y)
 ! CHECK:             %[[VAL_12:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_13:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_14:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@neqv_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
+! CHECK:             omp.wsloop reduction(byref @neqv_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_16:.*]]) : i32 = (%[[VAL_12]]) to (%[[VAL_13]]) inclusive step (%[[VAL_14]]) {
 ! CHECK:                 %[[VAL_17:.*]]:2 = hlfir.declare %[[VAL_15]] {uniq_name = "_QFsimple_reduction_switch_orderEx"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
 ! CHECK:                 fir.store %[[VAL_16]] to %[[VAL_11]]#1 : !fir.ref<i32>
@@ -153,7 +153,7 @@ subroutine simple_reduction_switch_order(y)
 ! CHECK:             %[[VAL_20:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_21:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_22:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@neqv_reduction %[[VAL_7]]#0 -> %[[VAL_23:.*]] : !fir.ref<!fir.logical<4>>, @neqv_reduction %[[VAL_9]]#0 -> %[[VAL_24:.*]] : !fir.ref<!fir.logical<4>>, @neqv_reduction %[[VAL_11]]#0 -> %[[VAL_25:.*]] : !fir.ref<!fir.logical<4>>) {
+! CHECK:             omp.wsloop reduction(byref @neqv_reduction %[[VAL_7]]#0 -> %[[VAL_23:.*]] : !fir.ref<!fir.logical<4>>, byref @neqv_reduction %[[VAL_9]]#0 -> %[[VAL_24:.*]] : !fir.ref<!fir.logical<4>>, byref @neqv_reduction %[[VAL_11]]#0 -> %[[VAL_25:.*]] : !fir.ref<!fir.logical<4>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_26:.*]]) : i32 = (%[[VAL_20]]) to (%[[VAL_21]]) inclusive step (%[[VAL_22]]) {
 ! CHECK:                 %[[VAL_27:.*]]:2 = hlfir.declare %[[VAL_23]] {uniq_name = "_QFmultiple_reductionsEx"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
 ! CHECK:                 %[[VAL_28:.*]]:2 = hlfir.declare %[[VAL_24]] {uniq_name = "_QFmultiple_reductionsEy"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-logical-or-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-logical-or-byref.f90
index 0af9e0d5c9fd2..1e5f91de801c4 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-logical-or-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-logical-or-byref.f90
@@ -41,7 +41,7 @@
 ! CHECK:             %[[VAL_12:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_13:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_14:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@or_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
+! CHECK:             omp.wsloop reduction(byref @or_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_16:.*]]) : i32 = (%[[VAL_12]]) to (%[[VAL_13]]) inclusive step (%[[VAL_14]]) {
 ! CHECK:                 %[[VAL_17:.*]]:2 = hlfir.declare %[[VAL_15]] {uniq_name = "_QFsimple_reductionEx"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
 ! CHECK:                 fir.store %[[VAL_16]] to %[[VAL_11]]#1 : !fir.ref<i32>
@@ -90,7 +90,7 @@ subroutine simple_reduction(y)
 ! CHECK:             %[[VAL_12:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_13:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_14:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@or_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
+! CHECK:             omp.wsloop reduction(byref @or_reduction %[[VAL_4]]#0 -> %[[VAL_15:.*]] : !fir.ref<!fir.logical<4>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_16:.*]]) : i32 = (%[[VAL_12]]) to (%[[VAL_13]]) inclusive step (%[[VAL_14]]) {
 ! CHECK:                 %[[VAL_17:.*]]:2 = hlfir.declare %[[VAL_15]] {uniq_name = "_QFsimple_reduction_switch_orderEx"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
 ! CHECK:                 fir.store %[[VAL_16]] to %[[VAL_11]]#1 : !fir.ref<i32>
@@ -149,7 +149,7 @@ subroutine simple_reduction_switch_order(y)
 ! CHECK:             %[[VAL_20:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_21:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_22:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@or_reduction %[[VAL_7]]#0 -> %[[VAL_23:.*]] : !fir.ref<!fir.logical<4>>, @or_reduction %[[VAL_9]]#0 -> %[[VAL_24:.*]] : !fir.ref<!fir.logical<4>>, @or_reduction %[[VAL_11]]#0 -> %[[VAL_25:.*]] : !fir.ref<!fir.logical<4>>) {
+! CHECK:             omp.wsloop reduction(byref @or_reduction %[[VAL_7]]#0 -> %[[VAL_23:.*]] : !fir.ref<!fir.logical<4>>, byref @or_reduction %[[VAL_9]]#0 -> %[[VAL_24:.*]] : !fir.ref<!fir.logical<4>>, byref @or_reduction %[[VAL_11]]#0 -> %[[VAL_25:.*]] : !fir.ref<!fir.logical<4>>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_26:.*]]) : i32 = (%[[VAL_20]]) to (%[[VAL_21]]) inclusive step (%[[VAL_22]]) {
 ! CHECK:                 %[[VAL_27:.*]]:2 = hlfir.declare %[[VAL_23]] {uniq_name = "_QFmultiple_reductionsEx"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
 ! CHECK:                 %[[VAL_28:.*]]:2 = hlfir.declare %[[VAL_24]] {uniq_name = "_QFmultiple_reductionsEy"} : (!fir.ref<!fir.logical<4>>) -> (!fir.ref<!fir.logical<4>>, !fir.ref<!fir.logical<4>>)
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-max-2-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-max-2-byref.f90
index 5358806cdcdec..28c70899e6ccc 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-max-2-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-max-2-byref.f90
@@ -1,7 +1,7 @@
 ! RUN: bbc -emit-hlfir -fopenmp --force-byref-reduction -o - %s 2>&1 | FileCheck %s
 ! RUN: %flang_fc1 -emit-hlfir -fopenmp -mmlir --force-byref-reduction -o - %s 2>&1 | FileCheck %s
 
-! CHECK: omp.wsloop byref reduction(@max_byref_i32
+! CHECK: omp.wsloop reduction(byref @max_byref_i32
 ! CHECK: arith.cmpi sgt
 ! CHECK: arith.select
 
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-max-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-max-byref.f90
index 11d039f9226cc..399afa1065560 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-max-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-max-byref.f90
@@ -46,7 +46,7 @@
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_10:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_11:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@max_byref_i32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<i32>) {
+! CHECK:             omp.wsloop reduction(byref @max_byref_i32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<i32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_13:.*]]) : i32 = (%[[VAL_9]]) to (%[[VAL_10]]) inclusive step (%[[VAL_11]]) {
 ! CHECK:                 %[[VAL_14:.*]]:2 = hlfir.declare %[[VAL_12]] {uniq_name = "_QFreduction_max_intEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 fir.store %[[VAL_13]] to %[[VAL_8]]#1 : !fir.ref<i32>
@@ -77,7 +77,7 @@
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_10:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_11:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@max_byref_f32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<f32>) {
+! CHECK:             omp.wsloop reduction(byref @max_byref_f32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<f32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_13:.*]]) : i32 = (%[[VAL_9]]) to (%[[VAL_10]]) inclusive step (%[[VAL_11]]) {
 ! CHECK:                 %[[VAL_14:.*]]:2 = hlfir.declare %[[VAL_12]] {uniq_name = "_QFreduction_max_realEx"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 ! CHECK:                 fir.store %[[VAL_13]] to %[[VAL_8]]#1 : !fir.ref<i32>
@@ -98,7 +98,7 @@
 ! CHECK:             %[[VAL_32:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_33:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_34:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@max_byref_f32 %[[VAL_4]]#0 -> %[[VAL_35:.*]] : !fir.ref<f32>) {
+! CHECK:             omp.wsloop reduction(byref @max_byref_f32 %[[VAL_4]]#0 -> %[[VAL_35:.*]] : !fir.ref<f32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_36:.*]]) : i32 = (%[[VAL_32]]) to (%[[VAL_33]]) inclusive step (%[[VAL_34]]) {
 ! CHECK:                 %[[VAL_37:.*]]:2 = hlfir.declare %[[VAL_35]] {uniq_name = "_QFreduction_max_realEx"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 ! CHECK:                 fir.store %[[VAL_36]] to %[[VAL_31]]#1 : !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-max-hlfir-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-max-hlfir-byref.f90
index a352cb195c25d..2f3aee9f1300a 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-max-hlfir-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-max-hlfir-byref.f90
@@ -33,7 +33,7 @@
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_10:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_11:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@max_byref_i32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<i32>) {
+! CHECK:             omp.wsloop reduction(byref @max_byref_i32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<i32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_13:.*]]) : i32 = (%[[VAL_9]]) to (%[[VAL_10]]) inclusive step (%[[VAL_11]]) {
 ! CHECK:                 %[[VAL_14:.*]]:2 = hlfir.declare %[[VAL_12]] {uniq_name = "_QFreduction_max_intEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 fir.store %[[VAL_13]] to %[[VAL_8]]#1 : !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-min-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-min-byref.f90
index d168b2a89295f..ccab017a1b0c6 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-min-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-min-byref.f90
@@ -46,7 +46,7 @@
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_10:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_11:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@min_byref_i32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<i32>) {
+! CHECK:             omp.wsloop reduction(byref @min_byref_i32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<i32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_13:.*]]) : i32 = (%[[VAL_9]]) to (%[[VAL_10]]) inclusive step (%[[VAL_11]]) {
 ! CHECK:                 %[[VAL_14:.*]]:2 = hlfir.declare %[[VAL_12]] {uniq_name = "_QFreduction_min_intEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 fir.store %[[VAL_13]] to %[[VAL_8]]#1 : !fir.ref<i32>
@@ -77,7 +77,7 @@
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_10:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_11:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@min_byref_f32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<f32>) {
+! CHECK:             omp.wsloop reduction(byref @min_byref_f32 %[[VAL_4]]#0 -> %[[VAL_12:.*]] : !fir.ref<f32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_13:.*]]) : i32 = (%[[VAL_9]]) to (%[[VAL_10]]) inclusive step (%[[VAL_11]]) {
 ! CHECK:                 %[[VAL_14:.*]]:2 = hlfir.declare %[[VAL_12]] {uniq_name = "_QFreduction_min_realEx"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 ! CHECK:                 fir.store %[[VAL_13]] to %[[VAL_8]]#1 : !fir.ref<i32>
@@ -100,7 +100,7 @@
 ! CHECK:             %[[VAL_32:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_33:.*]] = arith.constant 100 : i32
 ! CHECK:             %[[VAL_34:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@min_byref_f32 %[[VAL_4]]#0 -> %[[VAL_35:.*]] : !fir.ref<f32>) {
+! CHECK:             omp.wsloop reduction(byref @min_byref_f32 %[[VAL_4]]#0 -> %[[VAL_35:.*]] : !fir.ref<f32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_36:.*]]) : i32 = (%[[VAL_32]]) to (%[[VAL_33]]) inclusive step (%[[VAL_34]]) {
 ! CHECK:                 %[[VAL_37:.*]]:2 = hlfir.declare %[[VAL_35]] {uniq_name = "_QFreduction_min_realEx"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 ! CHECK:                 fir.store %[[VAL_36]] to %[[VAL_31]]#1 : !fir.ref<i32>
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-mul-byref.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-mul-byref.f90
index 7c538cdd470f8..762d9dd4f254f 100644
--- a/flang/test/Lower/OpenMP/wsloop-reduction-mul-byref.f90
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-mul-byref.f90
@@ -85,7 +85,7 @@
 ! CHECK:             %[[VAL_7:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_8:.*]] = arith.constant 10 : i32
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@multiply_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<i32>) {
+! CHECK:             omp.wsloop reduction(byref @multiply_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<i32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_11:.*]]) : i32 = (%[[VAL_7]]) to (%[[VAL_8]]) inclusive step (%[[VAL_9]]) {
 ! CHECK:                 %[[VAL_12:.*]]:2 = hlfir.declare %[[VAL_10]] {uniq_name = "_QFsimple_int_reductionEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 fir.store %[[VAL_11]] to %[[VAL_6]]#1 : !fir.ref<i32>
@@ -123,7 +123,7 @@ subroutine simple_int_reduction
 ! CHECK:             %[[VAL_7:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_8:.*]] = arith.constant 10 : i32
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@multiply_reduction_byref_f32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<f32>) {
+! CHECK:             omp.wsloop reduction(byref @multiply_reduction_byref_f32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<f32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_11:.*]]) : i32 = (%[[VAL_7]]) to (%[[VAL_8]]) inclusive step (%[[VAL_9]]) {
 ! CHECK:                 %[[VAL_12:.*]]:2 = hlfir.declare %[[VAL_10]] {uniq_name = "_QFsimple_real_reductionEx"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 ! CHECK:                 fir.store %[[VAL_11]] to %[[VAL_6]]#1 : !fir.ref<i32>
@@ -162,7 +162,7 @@ subroutine simple_real_reduction
 ! CHECK:             %[[VAL_7:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_8:.*]] = arith.constant 10 : i32
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@multiply_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<i32>) {
+! CHECK:             omp.wsloop reduction(byref @multiply_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<i32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_11:.*]]) : i32 = (%[[VAL_7]]) to (%[[VAL_8]]) inclusive step (%[[VAL_9]]) {
 ! CHECK:                 %[[VAL_12:.*]]:2 = hlfir.declare %[[VAL_10]] {uniq_name = "_QFsimple_int_reduction_switch_orderEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 fir.store %[[VAL_11]] to %[[VAL_6]]#1 : !fir.ref<i32>
@@ -200,7 +200,7 @@ subroutine simple_int_reduction_switch_order
 ! CHECK:             %[[VAL_7:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_8:.*]] = arith.constant 10 : i32
 ! CHECK:             %[[VAL_9:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@multiply_reduction_byref_f32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<f32>) {
+! CHECK:             omp.wsloop reduction(byref @multiply_reduction_byref_f32 %[[VAL_3]]#0 -> %[[VAL_10:.*]] : !fir.ref<f32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_11:.*]]) : i32 = (%[[VAL_7]]) to (%[[VAL_8]]) inclusive step (%[[VAL_9]]) {
 ! CHECK:                 %[[VAL_12:.*]]:2 = hlfir.declare %[[VAL_10]] {uniq_name = "_QFsimple_real_reduction_switch_orderEx"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 ! CHECK:                 fir.store %[[VAL_11]] to %[[VAL_6]]#1 : !fir.ref<i32>
@@ -247,7 +247,7 @@ subroutine simple_real_reduction_switch_order
 ! CHECK:             %[[VAL_13:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_14:.*]] = arith.constant 10 : i32
 ! CHECK:             %[[VAL_15:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@multiply_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_16:.*]] : !fir.ref<i32>, @multiply_reduction_byref_i32 %[[VAL_5]]#0 -> %[[VAL_17:.*]] : !fir.ref<i32>, @multiply_reduction_byref_i32 %[[VAL_7]]#0 -> %[[VAL_18:.*]] : !fir.ref<i32>) {
+! CHECK:             omp.wsloop reduction(byref @multiply_reduction_byref_i32 %[[VAL_3]]#0 -> %[[VAL_16:.*]] : !fir.ref<i32>, byref @multiply_reduction_byref_i32 %[[VAL_5]]#0 -> %[[VAL_17:.*]] : !fir.ref<i32>, byref @multiply_reduction_byref_i32 %[[VAL_7]]#0 -> %[[VAL_18:.*]] : !fir.ref<i32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_19:.*]]) : i32 = (%[[VAL_13]]) to (%[[VAL_14]]) inclusive step (%[[VAL_15]]) {
 ! CHECK:                 %[[VAL_20:.*]]:2 = hlfir.declare %[[VAL_16]] {uniq_name = "_QFmultiple_int_reductions_same_typeEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 %[[VAL_21:.*]]:2 = hlfir.declare %[[VAL_17]] {uniq_name = "_QFmultiple_int_reductions_same_typeEy"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
@@ -307,7 +307,7 @@ subroutine multiple_int_reductions_same_type
 ! CHECK:             %[[VAL_13:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_14:.*]] = arith.constant 10 : i32
 ! CHECK:             %[[VAL_15:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@multiply_reduction_byref_f32 %[[VAL_3]]#0 -> %[[VAL_16:.*]] : !fir.ref<f32>, @multiply_reduction_byref_f32 %[[VAL_5]]#0 -> %[[VAL_17:.*]] : !fir.ref<f32>, @multiply_reduction_byref_f32 %[[VAL_7]]#0 -> %[[VAL_18:.*]] : !fir.ref<f32>) {
+! CHECK:             omp.wsloop reduction(byref @multiply_reduction_byref_f32 %[[VAL_3]]#0 -> %[[VAL_16:.*]] : !fir.ref<f32>, byref @multiply_reduction_byref_f32 %[[VAL_5]]#0 -> %[[VAL_17:.*]] : !fir.ref<f32>, byref @multiply_reduction_byref_f32 %[[VAL_7]]#0 -> %[[VAL_18:.*]] : !fir.ref<f32>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_19:.*]]) : i32 = (%[[VAL_13]]) to (%[[VAL_14]]) inclusive step (%[[VAL_15]]) {
 ! CHECK:                 %[[VAL_20:.*]]:2 = hlfir.declare %[[VAL_16]] {uniq_name = "_QFmultiple_real_reductions_same_typeEx"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
 ! CHECK:                 %[[VAL_21:.*]]:2 = hlfir.declare %[[VAL_17]] {uniq_name = "_QFmultiple_real_reductions_same_typeEy"} : (!fir.ref<f32>) -> (!fir.ref<f32>, !fir.ref<f32>)
@@ -374,7 +374,7 @@ subroutine multiple_real_reductions_same_type
 ! CHECK:             %[[VAL_16:.*]] = arith.constant 1 : i32
 ! CHECK:             %[[VAL_17:.*]] = arith.constant 10 : i32
 ! CHECK:             %[[VAL_18:.*]] = arith.constant 1 : i32
-! CHECK:             omp.wsloop byref reduction(@multiply_reduction_byref_i32 %[[VAL_5]]#0 -> %[[VAL_19:.*]] : !fir.ref<i32>, @multiply_reduction_byref_i64 %[[VAL_7]]#0 -> %[[VAL_20:.*]] : !fir.ref<i64>, @multiply_reduction_byref_f32 %[[VAL_9]]#0 -> %[[VAL_21:.*]] : !fir.ref<f32>, @multiply_reduction_byref_f64 %[[VAL_3]]#0 -> %[[VAL_22:.*]] : !fir.ref<f64>) {
+! CHECK:             omp.wsloop reduction(byref @multiply_reduction_byref_i32 %[[VAL_5]]#0 -> %[[VAL_19:.*]] : !fir.ref<i32>, byref @multiply_reduction_byref_i64 %[[VAL_7]]#0 -> %[[VAL_20:.*]] : !fir.ref<i64>, byref @multiply_reduction_byref_f32 %[[VAL_9]]#0 -> %[[VAL_21:.*]] : !fir.ref<f32>, byref @multiply_reduction_byref_f64 %[[VAL_3]]#0 -> %[[VAL_22:.*]] : !fir.ref<f64>) {
 ! CHECK-NEXT:          omp.loop_nest (%[[VAL_23:.*]]) : i32 = (%[[VAL_16]]) to (%[[VAL_17]]) inclusive step (%[[VAL_18]]) {
 ! CHECK:                 %[[VAL_24:.*]]:2 = hlfir.declare %[[VAL_19]] {uniq_name = "_QFmultiple_reductions_different_typeEx"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
 ! CHECK:                 %[[VAL_25:.*]]:2 = hlfir.declare %[[VAL_20]] {uniq_name = "_QFmultiple_reductions_different_typeEy"} : (!fir.ref<i64>) -> (!fir.ref<i64>, !fir.ref<i64>)
diff --git a/flang/test/Lower/OpenMP/wsloop-reduction-multiple-clauses.f90 b/flang/test/Lower/OpenMP/wsloop-reduction-multiple-clauses.f90
new file mode 100644
index 0000000000000..4ae11c746e478
--- /dev/null
+++ b/flang/test/Lower/OpenMP/wsloop-reduction-multiple-clauses.f90
@@ -0,0 +1,164 @@
+! RUN: bbc -emit-hlfir -fopenmp -o %s | FileCheck %s
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -o - %s | FileCheck %s
+
+program main
+  implicit none
+
+  integer:: i
+  REAL(8) :: scalar
+  REAL(8) :: array(3,3)
+
+  scalar = 0
+  array = 0
+
+  !$omp parallel do reduction(+:scalar) reduction(+:array)
+  do i=1,10
+    scalar = i
+    array(1, 1) = i + 1
+    array(2, 2) = i + 2
+    array(3, 3) = i + 3
+  enddo
+
+  print *,scalar
+  print *,array
+
+endprogram
+
+! CHECK-LABEL:   omp.declare_reduction @add_reduction_byref_box_3x3xf64 : !fir.ref<!fir.box<!fir.array<3x3xf64>>> init {
+! CHECK:         ^bb0(%[[VAL_0:.*]]: !fir.ref<!fir.box<!fir.array<3x3xf64>>>):
+! CHECK:           %[[VAL_1:.*]] = arith.constant 0.000000e+00 : f64
+! CHECK:           %[[VAL_2:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.array<3x3xf64>>>
+! CHECK:           %[[VAL_3:.*]] = fir.alloca !fir.box<!fir.array<3x3xf64>>
+! CHECK:           %[[VAL_4:.*]] = arith.constant 3 : index
+! CHECK:           %[[VAL_5:.*]] = arith.constant 3 : index
+! CHECK:           %[[VAL_6:.*]] = fir.shape %[[VAL_4]], %[[VAL_5]] : (index, index) -> !fir.shape<2>
+! CHECK:           %[[VAL_7:.*]] = fir.allocmem !fir.array<3x3xf64> {bindc_name = ".tmp", uniq_name = ""}
+! CHECK:           %[[VAL_8:.*]] = arith.constant true
+! CHECK:           %[[VAL_9:.*]]:2 = hlfir.declare %[[VAL_7]](%[[VAL_6]]) {uniq_name = ".tmp"} : (!fir.heap<!fir.array<3x3xf64>>, !fir.shape<2>) -> (!fir.heap<!fir.array<3x3xf64>>, !fir.heap<!fir.array<3x3xf64>>)
+! CHECK:           %[[VAL_10:.*]] = arith.constant 0 : index
+! CHECK:           %[[VAL_11:.*]]:3 = fir.box_dims %[[VAL_2]], %[[VAL_10]] : (!fir.box<!fir.array<3x3xf64>>, index) -> (index, index, index)
+! CHECK:           %[[VAL_12:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_13:.*]]:3 = fir.box_dims %[[VAL_2]], %[[VAL_12]] : (!fir.box<!fir.array<3x3xf64>>, index) -> (index, index, index)
+! CHECK:           %[[VAL_14:.*]] = fir.shape_shift %[[VAL_11]]#0, %[[VAL_11]]#1, %[[VAL_13]]#0, %[[VAL_13]]#1 : (index, index, index, index) -> !fir.shapeshift<2>
+! CHECK:           %[[VAL_15:.*]] = fir.embox %[[VAL_9]]#0(%[[VAL_14]]) : (!fir.heap<!fir.array<3x3xf64>>, !fir.shapeshift<2>) -> !fir.box<!fir.array<3x3xf64>>
+! CHECK:           hlfir.assign %[[VAL_1]] to %[[VAL_15]] : f64, !fir.box<!fir.array<3x3xf64>>
+! CHECK:           fir.store %[[VAL_15]] to %[[VAL_3]] : !fir.ref<!fir.box<!fir.array<3x3xf64>>>
+! CHECK:           omp.yield(%[[VAL_3]] : !fir.ref<!fir.box<!fir.array<3x3xf64>>>)
+
+! CHECK-LABEL:   } combiner {
+! CHECK:         ^bb0(%[[VAL_0:.*]]: !fir.ref<!fir.box<!fir.array<3x3xf64>>>, %[[VAL_1:.*]]: !fir.ref<!fir.box<!fir.array<3x3xf64>>>):
+! CHECK:           %[[VAL_2:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.array<3x3xf64>>>
+! CHECK:           %[[VAL_3:.*]] = fir.load %[[VAL_1]] : !fir.ref<!fir.box<!fir.array<3x3xf64>>>
+! CHECK:           %[[VAL_4:.*]] = arith.constant 0 : index
+! CHECK:           %[[VAL_5:.*]]:3 = fir.box_dims %[[VAL_2]], %[[VAL_4]] : (!fir.box<!fir.array<3x3xf64>>, index) -> (index, index, index)
+! CHECK:           %[[VAL_6:.*]] = arith.constant 1 : index
+! CHECK:           %[[VAL_7:.*]]:3 = fir.box_dims %[[VAL_2]], %[[VAL_6]] : (!fir.box<!fir.array<3x3xf64>>, index) -> (index, index, index)
+! CHECK:           %[[VAL_8:.*]] = fir.shape_shift %[[VAL_5]]#0, %[[VAL_5]]#1, %[[VAL_7]]#0, %[[VAL_7]]#1 : (index, index, index, index) -> !fir.shapeshift<2>
+! CHECK:           %[[VAL_9:.*]] = arith.constant 1 : index
+! CHECK:           fir.do_loop %[[VAL_10:.*]] = %[[VAL_9]] to %[[VAL_7]]#1 step %[[VAL_9]] unordered {
+! CHECK:             fir.do_loop %[[VAL_11:.*]] = %[[VAL_9]] to %[[VAL_5]]#1 step %[[VAL_9]] unordered {
+! CHECK:               %[[VAL_12:.*]] = fir.array_coor %[[VAL_2]](%[[VAL_8]]) %[[VAL_11]], %[[VAL_10]] : (!fir.box<!fir.array<3x3xf64>>, !fir.shapeshift<2>, index, index) -> !fir.ref<f64>
+! CHECK:               %[[VAL_13:.*]] = fir.array_coor %[[VAL_3]](%[[VAL_8]]) %[[VAL_11]], %[[VAL_10]] : (!fir.box<!fir.array<3x3xf64>>, !fir.shapeshift<2>, index, index) -> !fir.ref<f64>
+! CHECK:               %[[VAL_14:.*]] = fir.load %[[VAL_12]] : !fir.ref<f64>
+! CHECK:               %[[VAL_15:.*]] = fir.load %[[VAL_13]] : !fir.ref<f64>
+! CHECK:               %[[VAL_16:.*]] = arith.addf %[[VAL_14]], %[[VAL_15]] fastmath<contract> : f64
+! CHECK:               fir.store %[[VAL_16]] to %[[VAL_12]] : !fir.ref<f64>
+! CHECK:             }
+! CHECK:           }
+! CHECK:           omp.yield(%[[VAL_0]] : !fir.ref<!fir.box<!fir.array<3x3xf64>>>)
+
+! CHECK-LABEL:   }  cleanup {
+! CHECK:         ^bb0(%[[VAL_0:.*]]: !fir.ref<!fir.box<!fir.array<3x3xf64>>>):
+! CHECK:           %[[VAL_1:.*]] = fir.load %[[VAL_0]] : !fir.ref<!fir.box<!fir.array<3x3xf64>>>
+! CHECK:           %[[VAL_2:.*]] = fir.box_addr %[[VAL_1]] : (!fir.box<!fir.array<3x3xf64>>) -> !fir.ref<!fir.array<3x3xf64>>
+! CHECK:           %[[VAL_3:.*]] = fir.convert %[[VAL_2]] : (!fir.ref<!fir.array<3x3xf64>>) -> i64
+! CHECK:           %[[VAL_4:.*]] = arith.constant 0 : i64
+! CHECK:           %[[VAL_5:.*]] = arith.cmpi ne, %[[VAL_3]], %[[VAL_4]] : i64
+! CHECK:           fir.if %[[VAL_5]] {
+! CHECK:             %[[VAL_6:.*]] = fir.convert %[[VAL_2]] : (!fir.ref<!fir.array<3x3xf64>>) -> !fir.heap<!fir.array<3x3xf64>>
+! CHECK:             fir.freemem %[[VAL_6]] : !fir.heap<!fir.array<3x3xf64>>
+! CHECK:           }
+! CHECK:           omp.yield
+! CHECK:         }
+
+! CHECK-LABEL:   omp.declare_reduction @add_reduction_f64 : f64 init {
+! CHECK:         ^bb0(%[[VAL_0:.*]]: f64):
+! CHECK:           %[[VAL_1:.*]] = arith.constant 0.000000e+00 : f64
+! CHECK:           omp.yield(%[[VAL_1]] : f64)
+
+! CHECK-LABEL:   } combiner {
+! CHECK:         ^bb0(%[[VAL_0:.*]]: f64, %[[VAL_1:.*]]: f64):
+! CHECK:           %[[VAL_2:.*]] = arith.addf %[[VAL_0]], %[[VAL_1]] fastmath<contract> : f64
+! CHECK:           omp.yield(%[[VAL_2]] : f64)
+! CHECK:         }
+
+! CHECK-LABEL:   func.func @_QQmain() attributes {fir.bindc_name = "main"} {
+! CHECK:           %[[VAL_0:.*]] = fir.address_of(@_QFEarray) : !fir.ref<!fir.array<3x3xf64>>
+! CHECK:           %[[VAL_1:.*]] = arith.constant 3 : index
+! CHECK:           %[[VAL_2:.*]] = arith.constant 3 : index
+! CHECK:           %[[VAL_3:.*]] = fir.shape %[[VAL_1]], %[[VAL_2]] : (index, index) -> !fir.shape<2>
+! CHECK:           %[[VAL_4:.*]]:2 = hlfir.declare %[[VAL_0]](%[[VAL_3]]) {uniq_name = "_QFEarray"} : (!fir.ref<!fir.array<3x3xf64>>, !fir.shape<2>) -> (!fir.ref<!fir.array<3x3xf64>>, !fir.ref<!fir.array<3x3xf64>>)
+! CHECK:           %[[VAL_5:.*]] = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFEi"}
+! CHECK:           %[[VAL_6:.*]]:2 = hlfir.declare %[[VAL_5]] {uniq_name = "_QFEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK:           %[[VAL_7:.*]] = fir.alloca f64 {bindc_name = "scalar", uniq_name = "_QFEscalar"}
+! CHECK:           %[[VAL_8:.*]]:2 = hlfir.declare %[[VAL_7]] {uniq_name = "_QFEscalar"} : (!fir.ref<f64>) -> (!fir.ref<f64>, !fir.ref<f64>)
+! CHECK:           %[[VAL_9:.*]] = arith.constant 0.000000e+00 : f64
+! CHECK:           hlfir.assign %[[VAL_9]] to %[[VAL_8]]#0 : f64, !fir.ref<f64>
+! CHECK:           %[[VAL_10:.*]] = arith.constant 0.000000e+00 : f64
+! CHECK:           hlfir.assign %[[VAL_10]] to %[[VAL_4]]#0 : f64, !fir.ref<!fir.array<3x3xf64>>
+! CHECK:           omp.parallel {
+! CHECK:             %[[VAL_11:.*]] = fir.alloca i32 {adapt.valuebyref, pinned}
+! CHECK:             %[[VAL_12:.*]]:2 = hlfir.declare %[[VAL_11]] {uniq_name = "_QFEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK:             %[[VAL_13:.*]] = arith.constant 1 : i32
+! CHECK:             %[[VAL_14:.*]] = arith.constant 10 : i32
+! CHECK:             %[[VAL_15:.*]] = arith.constant 1 : i32
+! CHECK:             %[[VAL_16:.*]] = fir.embox %[[VAL_4]]#0(%[[VAL_3]]) : (!fir.ref<!fir.array<3x3xf64>>, !fir.shape<2>) -> !fir.box<!fir.array<3x3xf64>>
+! CHECK:             %[[VAL_17:.*]] = fir.alloca !fir.box<!fir.array<3x3xf64>>
+! CHECK:             fir.store %[[VAL_16]] to %[[VAL_17]] : !fir.ref<!fir.box<!fir.array<3x3xf64>>>
+! CHECK:             omp.wsloop reduction(@add_reduction_f64 %[[VAL_8]]#0 -> %[[VAL_18:.*]] : !fir.ref<f64>, byref @add_reduction_byref_box_3x3xf64 %[[VAL_17]] -> %[[VAL_19:.*]] : !fir.ref<!fir.box<!fir.array<3x3xf64>>>) {
+! CHECK:               omp.loop_nest (%[[VAL_20:.*]]) : i32 = (%[[VAL_13]]) to (%[[VAL_14]]) inclusive step (%[[VAL_15]]) {
+! CHECK:                 %[[VAL_21:.*]]:2 = hlfir.declare %[[VAL_18]] {uniq_name = "_QFEscalar"} : (!fir.ref<f64>) -> (!fir.ref<f64>, !fir.ref<f64>)
+! CHECK:                 %[[VAL_22:.*]]:2 = hlfir.declare %[[VAL_19]] {uniq_name = "_QFEarray"} : (!fir.ref<!fir.box<!fir.array<3x3xf64>>>) -> (!fir.ref<!fir.box<!fir.array<3x3xf64>>>, !fir.ref<!fir.box<!fir.array<3x3xf64>>>)
+! CHECK:                 fir.store %[[VAL_20]] to %[[VAL_12]]#1 : !fir.ref<i32>
+! CHECK:                 %[[VAL_23:.*]] = fir.load %[[VAL_12]]#0 : !fir.ref<i32>
+! CHECK:                 %[[VAL_24:.*]] = fir.convert %[[VAL_23]] : (i32) -> f64
+! CHECK:                 hlfir.assign %[[VAL_24]] to %[[VAL_21]]#0 : f64, !fir.ref<f64>
+! CHECK:                 %[[VAL_25:.*]] = fir.load %[[VAL_12]]#0 : !fir.ref<i32>
+! CHECK:                 %[[VAL_26:.*]] = arith.constant 1 : i32
+! CHECK:                 %[[VAL_27:.*]] = arith.addi %[[VAL_25]], %[[VAL_26]] : i32
+! CHECK:                 %[[VAL_28:.*]] = fir.convert %[[VAL_27]] : (i32) -> f64
+! CHECK:                 %[[VAL_29:.*]] = fir.load %[[VAL_22]]#0 : !fir.ref<!fir.box<!fir.array<3x3xf64>>>
+! CHECK:                 %[[VAL_30:.*]] = arith.constant 1 : index
+! CHECK:                 %[[VAL_31:.*]] = arith.constant 1 : index
+! CHECK:                 %[[VAL_32:.*]] = hlfir.designate %[[VAL_29]] (%[[VAL_30]], %[[VAL_31]])  : (!fir.box<!fir.array<3x3xf64>>, index, index) -> !fir.ref<f64>
+! CHECK:                 hlfir.assign %[[VAL_28]] to %[[VAL_32]] : f64, !fir.ref<f64>
+! CHECK:                 %[[VAL_33:.*]] = fir.load %[[VAL_12]]#0 : !fir.ref<i32>
+! CHECK:                 %[[VAL_34:.*]] = arith.constant 2 : i32
+! CHECK:                 %[[VAL_35:.*]] = arith.addi %[[VAL_33]], %[[VAL_34]] : i32
+! CHECK:                 %[[VAL_36:.*]] = fir.convert %[[VAL_35]] : (i32) -> f64
+! CHECK:                 %[[VAL_37:.*]] = fir.load %[[VAL_22]]#0 : !fir.ref<!fir.box<!fir.array<3x3xf64>>>
+! CHECK:                 %[[VAL_38:.*]] = arith.constant 2 : index
+! CHECK:                 %[[VAL_39:.*]] = arith.constant 2 : index
+! CHECK:                 %[[VAL_40:.*]] = hlfir.designate %[[VAL_37]] (%[[VAL_38]], %[[VAL_39]])  : (!fir.box<!fir.array<3x3xf64>>, index, index) -> !fir.ref<f64>
+! CHECK:                 hlfir.assign %[[VAL_36]] to %[[VAL_40]] : f64, !fir.ref<f64>
+! CHECK:                 %[[VAL_41:.*]] = fir.load %[[VAL_12]]#0 : !fir.ref<i32>
+! CHECK:                 %[[VAL_42:.*]] = arith.constant 3 : i32
+! CHECK:                 %[[VAL_43:.*]] = arith.addi %[[VAL_41]], %[[VAL_42]] : i32
+! CHECK:                 %[[VAL_44:.*]] = fir.convert %[[VAL_43]] : (i32) -> f64
+! CHECK:                 %[[VAL_45:.*]] = fir.load %[[VAL_22]]#0 : !fir.ref<!fir.box<!fir.array<3x3xf64>>>
+! CHECK:                 %[[VAL_46:.*]] = arith.constant 3 : index
+! CHECK:                 %[[VAL_47:.*]] = arith.constant 3 : index
+! CHECK:                 %[[VAL_48:.*]] = hlfir.designate %[[VAL_45]] (%[[VAL_46]], %[[VAL_47]])  : (!fir.box<!fir.array<3x3xf64>>, index, index) -> !fir.ref<f64>
+! CHECK:                 hlfir.assign %[[VAL_44]] to %[[VAL_48]] : f64, !fir.ref<f64>
+! CHECK:                 omp.yield
+! CHECK:               }
+! CHECK:               omp.terminator
+! CHECK:             }
+! CHECK:             omp.terminator
+! CHECK:           }
+
+! CHECK-LABEL:   fir.global internal @_QFEarray : !fir.array<3x3xf64> {
+! CHECK:           %[[VAL_0:.*]] = fir.zero_bits !fir.array<3x3xf64>
+! CHECK:           fir.has_value %[[VAL_0]] : !fir.array<3x3xf64>
+! CHECK:         }
+! CHECK:         func.func private @_FortranAioBeginExternalListOutput(i32, !fir.ref<i8>, i32) -> !fir.ref<i8> attributes {fir.io, fir.runtime}
diff --git a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
index c9ee0c25194c2..e54ec4f2b1d72 100644
--- a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
+++ b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
@@ -1344,7 +1344,7 @@ class OpenMPIRBuilder {
   InsertPointTy createReductions(const LocationDescription &Loc,
                                  InsertPointTy AllocaIP,
                                  ArrayRef<ReductionInfo> ReductionInfos,
-                                 bool IsNoWait = false, bool IsByRef = false);
+                                 ArrayRef<bool> IsByRef, bool IsNoWait = false);
 
   ///}
 
diff --git a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
index 391a4947877a6..74dc825239034 100644
--- a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
+++ b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
@@ -2121,9 +2121,11 @@ Function *getFreshReductionFunc(Module &M) {
                           ".omp.reduction.func", &M);
 }
 
-OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::createReductions(
-    const LocationDescription &Loc, InsertPointTy AllocaIP,
-    ArrayRef<ReductionInfo> ReductionInfos, bool IsNoWait, bool IsByRef) {
+OpenMPIRBuilder::InsertPointTy
+OpenMPIRBuilder::createReductions(const LocationDescription &Loc,
+                                  InsertPointTy AllocaIP,
+                                  ArrayRef<ReductionInfo> ReductionInfos,
+                                  ArrayRef<bool> IsByRef, bool IsNoWait) {
   for (const ReductionInfo &RI : ReductionInfos) {
     (void)RI;
     assert(RI.Variable && "expected non-null variable");
@@ -2213,7 +2215,7 @@ OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::createReductions(
     // We have one less load for by-ref case because that load is now inside of
     // the reduction region
     Value *RedValue = nullptr;
-    if (!IsByRef) {
+    if (!IsByRef[En.index()]) {
       RedValue = Builder.CreateLoad(ValueType, RI.Variable,
                                     "red.value." + Twine(En.index()));
     }
@@ -2221,7 +2223,7 @@ OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::createReductions(
         Builder.CreateLoad(ValueType, RI.PrivateVariable,
                            "red.private.value." + Twine(En.index()));
     Value *Reduced;
-    if (IsByRef) {
+    if (IsByRef[En.index()]) {
       Builder.restoreIP(RI.ReductionGen(Builder.saveIP(), RI.Variable,
                                         PrivateRedValue, Reduced));
     } else {
@@ -2231,7 +2233,7 @@ OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::createReductions(
     if (!Builder.GetInsertBlock())
       return InsertPointTy();
     // for by-ref case, the load is inside of the reduction region
-    if (!IsByRef)
+    if (!IsByRef[En.index()])
       Builder.CreateStore(Reduced, RI.Variable);
   }
   Function *EndReduceFunc = getOrCreateRuntimeFunctionPtr(
@@ -2244,7 +2246,7 @@ OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::createReductions(
   // function. There are no loads/stores here because they will be happening
   // inside the atomic elementwise reduction.
   Builder.SetInsertPoint(AtomicRedBlock);
-  if (CanGenerateAtomic && !IsByRef) {
+  if (CanGenerateAtomic && llvm::none_of(IsByRef, [](bool P) { return P; })) {
     for (const ReductionInfo &RI : ReductionInfos) {
       Builder.restoreIP(RI.AtomicReductionGen(Builder.saveIP(), RI.ElementType,
                                               RI.Variable, RI.PrivateVariable));
@@ -2283,7 +2285,7 @@ OpenMPIRBuilder::InsertPointTy OpenMPIRBuilder::createReductions(
     if (!Builder.GetInsertBlock())
       return InsertPointTy();
     // store is inside of the reduction region when using by-ref
-    if (!IsByRef)
+    if (!IsByRef[En.index()])
       Builder.CreateStore(Reduced, LHSPtr);
   }
   Builder.CreateRetVoid();
diff --git a/llvm/unittests/Frontend/OpenMPIRBuilderTest.cpp b/llvm/unittests/Frontend/OpenMPIRBuilderTest.cpp
index 8344bca08404e..3ed3034f489ce 100644
--- a/llvm/unittests/Frontend/OpenMPIRBuilderTest.cpp
+++ b/llvm/unittests/Frontend/OpenMPIRBuilderTest.cpp
@@ -4982,8 +4982,10 @@ TEST_F(OpenMPIRBuilderTest, CreateReductions) {
   OpenMPIRBuilder::ReductionInfo ReductionInfos[] = {
       {SumType, SumReduced, SumPrivatized, sumReduction, sumAtomicReduction},
       {XorType, XorReduced, XorPrivatized, xorReduction, xorAtomicReduction}};
+  bool ReduceVariableByRef[] = {false, false};
 
-  OMPBuilder.createReductions(BodyIP, BodyAllocaIP, ReductionInfos);
+  OMPBuilder.createReductions(BodyIP, BodyAllocaIP, ReductionInfos,
+                              ReduceVariableByRef);
 
   Builder.restoreIP(AfterIP);
   Builder.CreateRetVoid();
@@ -5230,12 +5232,16 @@ TEST_F(OpenMPIRBuilderTest, CreateTwoReductions) {
       /* NumThreads */ nullptr, OMP_PROC_BIND_default,
       /* IsCancellable */ false);
 
+  bool ReduceVariableByRef[] = {false};
+
   OMPBuilder.createReductions(
       FirstBodyIP, FirstBodyAllocaIP,
-      {{SumType, SumReduced, SumPrivatized, sumReduction, sumAtomicReduction}});
+      {{SumType, SumReduced, SumPrivatized, sumReduction, sumAtomicReduction}},
+      ReduceVariableByRef);
   OMPBuilder.createReductions(
       SecondBodyIP, SecondBodyAllocaIP,
-      {{XorType, XorReduced, XorPrivatized, xorReduction, xorAtomicReduction}});
+      {{XorType, XorReduced, XorPrivatized, xorReduction, xorAtomicReduction}},
+      ReduceVariableByRef);
 
   Builder.restoreIP(AfterIP);
   Builder.CreateRetVoid();
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPClauseOperands.h b/mlir/include/mlir/Dialect/OpenMP/OpenMPClauseOperands.h
index 244cee1dd635b..de7959db489e9 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPClauseOperands.h
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPClauseOperands.h
@@ -177,8 +177,8 @@ struct ProcBindClauseOps {
 
 struct ReductionClauseOps {
   llvm::SmallVector<Value> reductionVars;
+  llvm::SmallVector<bool> reduceVarByRef;
   llvm::SmallVector<Attribute> reductionDeclSymbols;
-  UnitAttr reductionByRefAttr;
 };
 
 struct SafelenClauseOps {
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
index f248be1639fe9..bbd5ff18c3780 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
@@ -272,22 +272,21 @@ def ParallelOp : OpenMP_Op<"parallel", [
     that specify the memory allocator to be used to obtain storage for private values.
 
     Reductions can be performed in a parallel construct by specifying reduction
-    accumulator variables in `reduction_vars` and symbols referring to reduction
-    declarations in the `reductions` attribute. Each reduction is identified
-    by the accumulator it uses and accumulators must not be repeated in the same
-    reduction. The `omp.reduction` operation accepts the accumulator and a
-    partial value which is considered to be produced by the thread for the
-    given reduction. If multiple values are produced for the same accumulator,
-    i.e. there are multiple `omp.reduction`s, the last value is taken. The
-    reduction declaration specifies how to combine the values from each thread
-    into the final value, which is available in the accumulator after all the
-    threads complete.
+    accumulator variables in `reduction_vars`, symbols referring to reduction
+    declarations in the `reductions` attribute, and whether the reduction
+    variable should be passed into the redution region by value or by reference
+    in `reduction_vars_byref`. Each reduction is identified by the accumulator
+    it uses and accumulators must not be repeated in the same reduction. The
+    `omp.reduction` operation accepts the accumulator and a partial value which
+    is considered to be produced by the thread for the given reduction. If
+    multiple values are produced for the same accumulator, i.e. there are
+    multiple `omp.reduction`s, the last value is taken. The reduction
+    declaration specifies how to combine the values from each thread into the
+    final value, which is available in the accumulator after all the threads
+    complete.
 
     The optional $proc_bind_val attribute controls the thread affinity for the execution
     of the parallel region.
-
-    The optional byref attribute controls whether reduction arguments are passed by
-    reference or by value.
   }];
 
   let arguments = (ins Optional<I1>:$if_expr_var,
@@ -295,11 +294,11 @@ def ParallelOp : OpenMP_Op<"parallel", [
              Variadic<AnyType>:$allocate_vars,
              Variadic<AnyType>:$allocators_vars,
              Variadic<OpenMP_PointerLikeType>:$reduction_vars,
+             OptionalAttr<DenseBoolArrayAttr>:$reduction_vars_byref,
              OptionalAttr<SymbolRefArrayAttr>:$reductions,
              OptionalAttr<ProcBindKindAttr>:$proc_bind_val,
              Variadic<AnyType>:$private_vars,
-             OptionalAttr<SymbolRefArrayAttr>:$privatizers,
-             UnitAttr:$byref);
+             OptionalAttr<SymbolRefArrayAttr>:$privatizers);
 
   let regions = (region AnyRegion:$region);
 
@@ -321,10 +320,9 @@ def ParallelOp : OpenMP_Op<"parallel", [
                 $allocators_vars, type($allocators_vars)
               ) `)`
           | `proc_bind` `(` custom<ClauseAttr>($proc_bind_val) `)`
-          | `byref` $byref
     ) custom<ParallelRegion>($region, $reduction_vars, type($reduction_vars),
-                             $reductions, $private_vars, type($private_vars),
-                             $privatizers) attr-dict
+                             $reduction_vars_byref, $reductions, $private_vars,
+                             type($private_vars), $privatizers) attr-dict
   }];
   let hasVerifier = 1;
 }
@@ -649,14 +647,15 @@ def WsloopOp : OpenMP_Op<"wsloop", [AttrSizedOperandSegments,
     elements.
 
     Reductions can be performed in a worksharing-loop by specifying reduction
-    accumulator variables in `reduction_vars` and symbols referring to reduction
-    declarations in the `reductions` attribute. Each reduction is identified
-    by the accumulator it uses and accumulators must not be repeated in the same
-    reduction. A private variable corresponding to the accumulator is used in
-    place of the accumulator inside the body of the worksharing-loop. The
-    reduction declaration specifies how to combine the values from each
-    iteration into the final value, which is available in the accumulator after
-    the loop completes.
+    accumulator variables in `reduction_vars`, symbols referring to reduction
+    declarations in the `reductions` attribute, and whether the reduction
+    variable should be passed by reference or value in `reduction_vars_byref`.
+    Each reduction is identified by the accumulator it uses and accumulators
+    must not be repeated in the same reduction. A private variable corresponding
+    to the accumulator is used in place of the accumulator inside the body of
+    the worksharing-loop. The reduction declaration specifies how to combine the
+    values from each iteration into the final value, which is available in the
+    accumulator after the loop completes.
 
     The optional `schedule_val` attribute specifies the loop schedule for this
     loop, determining how the loop is distributed across the parallel threads.
@@ -677,21 +676,18 @@ def WsloopOp : OpenMP_Op<"wsloop", [AttrSizedOperandSegments,
     The optional `order` attribute specifies which order the iterations of the
     associate loops are executed in. Currently the only option for this
     attribute is "concurrent".
-
-    The optional `byref` attribute indicates that reduction arguments should be
-    passed by reference.
   }];
 
   let arguments = (ins Variadic<AnyType>:$linear_vars,
              Variadic<I32>:$linear_step_vars,
              Variadic<OpenMP_PointerLikeType>:$reduction_vars,
+             OptionalAttr<DenseBoolArrayAttr>:$reduction_vars_byref,
              OptionalAttr<SymbolRefArrayAttr>:$reductions,
              OptionalAttr<ScheduleKindAttr>:$schedule_val,
              Optional<AnyType>:$schedule_chunk_var,
              OptionalAttr<ScheduleModifierAttr>:$schedule_modifier,
              UnitAttr:$simd_modifier,
              UnitAttr:$nowait,
-             UnitAttr:$byref,
              ConfinedAttr<OptionalAttr<I64Attr>, [IntMinValue<0>]>:$ordered_val,
              OptionalAttr<OrderKindAttr>:$order_val);
 
@@ -716,11 +712,10 @@ def WsloopOp : OpenMP_Op<"wsloop", [AttrSizedOperandSegments,
                 $schedule_val, $schedule_modifier, $simd_modifier,
                 $schedule_chunk_var, type($schedule_chunk_var)) `)`
           |`nowait` $nowait
-          |`byref` $byref
           |`ordered` `(` $ordered_val `)`
           |`order` `(` custom<ClauseAttr>($order_val) `)`
     ) custom<Wsloop>($region, $reduction_vars, type($reduction_vars),
-                     $reductions) attr-dict
+                     $reduction_vars_byref, $reductions) attr-dict
   }];
   let hasVerifier = 1;
 }
diff --git a/mlir/lib/Conversion/SCFToOpenMP/SCFToOpenMP.cpp b/mlir/lib/Conversion/SCFToOpenMP/SCFToOpenMP.cpp
index d6f85451ee5d3..b4d92dd630aff 100644
--- a/mlir/lib/Conversion/SCFToOpenMP/SCFToOpenMP.cpp
+++ b/mlir/lib/Conversion/SCFToOpenMP/SCFToOpenMP.cpp
@@ -449,6 +449,7 @@ struct ParallelOpLowering : public OpRewritePattern<scf::ParallelOp> {
         /* allocate_vars = */ llvm::SmallVector<Value>{},
         /* allocators_vars = */ llvm::SmallVector<Value>{},
         /* reduction_vars = */ llvm::SmallVector<Value>{},
+        /* reduction_vars_isbyref = */ DenseBoolArrayAttr{},
         /* reductions = */ ArrayAttr{},
         /* proc_bind_val = */ omp::ClauseProcBindKindAttr{},
         /* private_vars = */ ValueRange(),
@@ -467,6 +468,10 @@ struct ParallelOpLowering : public OpRewritePattern<scf::ParallelOp> {
           wsloopOp.setReductionsAttr(
               ArrayAttr::get(rewriter.getContext(), reductionDeclSymbols));
           wsloopOp.getReductionVarsMutable().append(reductionVariables);
+          llvm::SmallVector<bool> byRefVec;
+          byRefVec.resize(reductionVariables.size(), false);
+          wsloopOp.setReductionVarsByref(
+              DenseBoolArrayAttr::get(rewriter.getContext(), byRefVec));
         }
         rewriter.create<omp::TerminatorOp>(loc); // omp.parallel terminator.
 
diff --git a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
index 61073af2aa4d6..a7902da68c4c8 100644
--- a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
+++ b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
@@ -15,11 +15,13 @@
 #include "mlir/Dialect/LLVMIR/LLVMTypes.h"
 #include "mlir/Dialect/OpenACCMPCommon/Interfaces/AtomicInterfaces.h"
 #include "mlir/IR/Attributes.h"
+#include "mlir/IR/BuiltinAttributes.h"
 #include "mlir/IR/DialectImplementation.h"
 #include "mlir/IR/OpImplementation.h"
 #include "mlir/IR/OperationSupport.h"
 #include "mlir/Interfaces/FoldInterfaces.h"
 
+#include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/BitVector.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/STLForwardCompat.h"
@@ -435,25 +437,30 @@ static void printScheduleClause(OpAsmPrinter &p, Operation *op,
 // Parser, printer and verifier for ReductionVarList
 //===----------------------------------------------------------------------===//
 
-ParseResult parseClauseWithRegionArgs(
+static ParseResult parseClauseWithRegionArgs(
     OpAsmParser &parser, Region &region,
     SmallVectorImpl<OpAsmParser::UnresolvedOperand> &operands,
-    SmallVectorImpl<Type> &types, ArrayAttr &symbols,
+    SmallVectorImpl<Type> &types, DenseBoolArrayAttr &isByRef,
+    ArrayAttr &symbols,
     SmallVectorImpl<OpAsmParser::Argument> &regionPrivateArgs) {
   SmallVector<SymbolRefAttr> reductionVec;
+  SmallVector<bool> isByRefVec;
   unsigned regionArgOffset = regionPrivateArgs.size();
 
   if (failed(
           parser.parseCommaSeparatedList(OpAsmParser::Delimiter::Paren, [&]() {
+            ParseResult optionalByref = parser.parseOptionalKeyword("byref");
             if (parser.parseAttribute(reductionVec.emplace_back()) ||
                 parser.parseOperand(operands.emplace_back()) ||
                 parser.parseArrow() ||
                 parser.parseArgument(regionPrivateArgs.emplace_back()) ||
                 parser.parseColonType(types.emplace_back()))
               return failure();
+            isByRefVec.push_back(optionalByref.succeeded());
             return success();
           })))
     return failure();
+  isByRef = DenseBoolArrayAttr::get(parser.getContext(), isByRefVec);
 
   auto *argsBegin = regionPrivateArgs.begin();
   MutableArrayRef argsSubrange(argsBegin + regionArgOffset,
@@ -469,15 +476,18 @@ ParseResult parseClauseWithRegionArgs(
 static void printClauseWithRegionArgs(OpAsmPrinter &p, Operation *op,
                                       ValueRange argsSubrange,
                                       StringRef clauseName, ValueRange operands,
-                                      TypeRange types, ArrayAttr symbols) {
+                                      TypeRange types, DenseBoolArrayAttr byRef,
+                                      ArrayAttr symbols) {
   if (!clauseName.empty())
     p << clauseName << "(";
 
-  llvm::interleaveComma(
-      llvm::zip_equal(symbols, operands, argsSubrange, types), p, [&p](auto t) {
-        auto [sym, op, arg, type] = t;
-        p << sym << " " << op << " -> " << arg << " : " << type;
-      });
+  llvm::interleaveComma(llvm::zip_equal(symbols, operands, argsSubrange, types,
+                                        byRef.asArrayRef()),
+                        p, [&p](auto t) {
+                          auto [sym, op, arg, type, isByRef] = t;
+                          p << (isByRef ? "byref " : "") << sym << " " << op
+                            << " -> " << arg << " : " << type;
+                        });
 
   if (!clauseName.empty())
     p << ") ";
@@ -486,7 +496,8 @@ static void printClauseWithRegionArgs(OpAsmPrinter &p, Operation *op,
 static ParseResult parseParallelRegion(
     OpAsmParser &parser, Region &region,
     SmallVectorImpl<OpAsmParser::UnresolvedOperand> &reductionVarOperands,
-    SmallVectorImpl<Type> &reductionVarTypes, ArrayAttr &reductionSymbols,
+    SmallVectorImpl<Type> &reductionVarTypes,
+    DenseBoolArrayAttr &reductionByRef, ArrayAttr &reductionSymbols,
     llvm::SmallVectorImpl<OpAsmParser::UnresolvedOperand> &privateVarOperands,
     llvm::SmallVectorImpl<Type> &privateVarsTypes,
     ArrayAttr &privatizerSymbols) {
@@ -494,16 +505,23 @@ static ParseResult parseParallelRegion(
 
   if (succeeded(parser.parseOptionalKeyword("reduction"))) {
     if (failed(parseClauseWithRegionArgs(parser, region, reductionVarOperands,
-                                         reductionVarTypes, reductionSymbols,
-                                         regionPrivateArgs)))
+                                         reductionVarTypes, reductionByRef,
+                                         reductionSymbols, regionPrivateArgs)))
       return failure();
   }
 
   if (succeeded(parser.parseOptionalKeyword("private"))) {
+    auto privateByRef = DenseBoolArrayAttr::get(parser.getContext(), {});
     if (failed(parseClauseWithRegionArgs(parser, region, privateVarOperands,
-                                         privateVarsTypes, privatizerSymbols,
-                                         regionPrivateArgs)))
+                                         privateVarsTypes, privateByRef,
+                                         privatizerSymbols, regionPrivateArgs)))
       return failure();
+    if (llvm::any_of(privateByRef.asArrayRef(),
+                     [](bool byref) { return byref; })) {
+      parser.emitError(parser.getCurrentLocation(),
+                       "private clause cannot have byref attributes");
+      return failure();
+    }
   }
 
   return parser.parseRegion(region, regionPrivateArgs);
@@ -512,6 +530,7 @@ static ParseResult parseParallelRegion(
 static void printParallelRegion(OpAsmPrinter &p, Operation *op, Region &region,
                                 ValueRange reductionVarOperands,
                                 TypeRange reductionVarTypes,
+                                DenseBoolArrayAttr reductionVarIsByRef,
                                 ArrayAttr reductionSymbols,
                                 ValueRange privateVarOperands,
                                 TypeRange privateVarTypes,
@@ -522,7 +541,7 @@ static void printParallelRegion(OpAsmPrinter &p, Operation *op, Region &region,
                                  argsBegin + reductionVarTypes.size());
     printClauseWithRegionArgs(p, op, argsSubrange, "reduction",
                               reductionVarOperands, reductionVarTypes,
-                              reductionSymbols);
+                              reductionVarIsByRef, reductionSymbols);
   }
 
   if (privatizerSymbols) {
@@ -530,8 +549,13 @@ static void printParallelRegion(OpAsmPrinter &p, Operation *op, Region &region,
     MutableArrayRef argsSubrange(argsBegin + reductionVarOperands.size(),
                                  argsBegin + reductionVarOperands.size() +
                                      privateVarTypes.size());
+    mlir::SmallVector<bool> isByRefVec;
+    isByRefVec.resize(privateVarTypes.size(), false);
+    DenseBoolArrayAttr isByRef =
+        DenseBoolArrayAttr::get(op->getContext(), isByRefVec);
+
     printClauseWithRegionArgs(p, op, argsSubrange, "private",
-                              privateVarOperands, privateVarTypes,
+                              privateVarOperands, privateVarTypes, isByRef,
                               privatizerSymbols);
   }
 
@@ -1163,9 +1187,14 @@ static void printPrivateList(OpAsmPrinter &p, Operation *op,
   MutableArrayRef argsSubrange(argsBegin + targetOp.getMapOperands().size(),
                                argsBegin + targetOp.getMapOperands().size() +
                                    privateVarTypes.size());
+  mlir::SmallVector<bool> isByRefVec;
+  isByRefVec.resize(privateVarTypes.size(), false);
+  DenseBoolArrayAttr isByRef =
+      DenseBoolArrayAttr::get(op->getContext(), isByRefVec);
+
   printClauseWithRegionArgs(
       p, op, argsSubrange, /*clauseName=*/llvm::StringRef{}, privateVarOperands,
-      privateVarTypes, privatizerSymbols);
+      privateVarTypes, isByRef, privatizerSymbols);
 }
 
 static void printCaptureType(OpAsmPrinter &p, Operation *op,
@@ -1402,21 +1431,24 @@ void ParallelOp::build(OpBuilder &builder, OperationState &state,
   ParallelOp::build(
       builder, state, /*if_expr_var=*/nullptr, /*num_threads_var=*/nullptr,
       /*allocate_vars=*/ValueRange(), /*allocators_vars=*/ValueRange(),
-      /*reduction_vars=*/ValueRange(), /*reductions=*/nullptr,
+      /*reduction_vars=*/ValueRange(), /*reduction_vars_byref=*/nullptr,
+      /*reductions=*/nullptr,
       /*proc_bind_val=*/nullptr, /*private_vars=*/ValueRange(),
-      /*privatizers=*/nullptr, /*byref=*/false);
+      /*privatizers=*/nullptr);
   state.addAttributes(attributes);
 }
 
 void ParallelOp::build(OpBuilder &builder, OperationState &state,
                        const ParallelClauseOps &clauses) {
   MLIRContext *ctx = builder.getContext();
-  ParallelOp::build(
-      builder, state, clauses.ifVar, clauses.numThreadsVar,
-      clauses.allocateVars, clauses.allocatorVars, clauses.reductionVars,
-      makeArrayAttr(ctx, clauses.reductionDeclSymbols),
-      clauses.procBindKindAttr, clauses.privateVars,
-      makeArrayAttr(ctx, clauses.privatizers), clauses.reductionByRefAttr);
+
+  ParallelOp::build(builder, state, clauses.ifVar, clauses.numThreadsVar,
+                    clauses.allocateVars, clauses.allocatorVars,
+                    clauses.reductionVars,
+                    DenseBoolArrayAttr::get(ctx, clauses.reduceVarByRef),
+                    makeArrayAttr(ctx, clauses.reductionDeclSymbols),
+                    clauses.procBindKindAttr, clauses.privateVars,
+                    makeArrayAttr(ctx, clauses.privatizers));
 }
 
 template <typename OpType>
@@ -1488,6 +1520,13 @@ LogicalResult ParallelOp::verify() {
   if (failed(verifyPrivateVarList(*this)))
     return failure();
 
+  auto reductionVarsByRef = getReductionVarsByref();
+  if (reductionVarsByRef &&
+      reductionVarsByRef->size() != getReductionVars().size())
+    return emitOpError()
+           << "expected as many reduction variable by reference attributes "
+              "as reduction variables";
+
   return verifyReductionVarList(*this, getReductions(), getReductionVars());
 }
 
@@ -1609,13 +1648,13 @@ ParseResult
 parseWsloop(OpAsmParser &parser, Region &region,
             SmallVectorImpl<OpAsmParser::UnresolvedOperand> &reductionOperands,
             SmallVectorImpl<Type> &reductionTypes,
-            ArrayAttr &reductionSymbols) {
+            DenseBoolArrayAttr &reductionByRef, ArrayAttr &reductionSymbols) {
   // Parse an optional reduction clause
   llvm::SmallVector<OpAsmParser::Argument> privates;
   if (succeeded(parser.parseOptionalKeyword("reduction"))) {
     if (failed(parseClauseWithRegionArgs(parser, region, reductionOperands,
-                                         reductionTypes, reductionSymbols,
-                                         privates)))
+                                         reductionTypes, reductionByRef,
+                                         reductionSymbols, privates)))
       return failure();
   }
   return parser.parseRegion(region, privates);
@@ -1623,11 +1662,11 @@ parseWsloop(OpAsmParser &parser, Region &region,
 
 void printWsloop(OpAsmPrinter &p, Operation *op, Region &region,
                  ValueRange reductionOperands, TypeRange reductionTypes,
-                 ArrayAttr reductionSymbols) {
+                 DenseBoolArrayAttr isByRef, ArrayAttr reductionSymbols) {
   if (reductionSymbols) {
     auto reductionArgs = region.front().getArguments();
     printClauseWithRegionArgs(p, op, reductionArgs, "reduction",
-                              reductionOperands, reductionTypes,
+                              reductionOperands, reductionTypes, isByRef,
                               reductionSymbols);
   }
   p.printRegion(region, /*printEntryBlockArgs=*/false);
@@ -1637,9 +1676,10 @@ void WsloopOp::build(OpBuilder &builder, OperationState &state,
                      ArrayRef<NamedAttribute> attributes) {
   build(builder, state, /*linear_vars=*/ValueRange(),
         /*linear_step_vars=*/ValueRange(), /*reduction_vars=*/ValueRange(),
+        /*reduction_vars_byref=*/nullptr,
         /*reductions=*/nullptr, /*schedule_val=*/nullptr,
         /*schedule_chunk_var=*/nullptr, /*schedule_modifier=*/nullptr,
-        /*simd_modifier=*/false, /*nowait=*/false, /*byref=*/false,
+        /*simd_modifier=*/false, /*nowait=*/false,
         /*ordered_val=*/nullptr, /*order_val=*/nullptr);
   state.addAttributes(attributes);
 }
@@ -1649,12 +1689,13 @@ void WsloopOp::build(OpBuilder &builder, OperationState &state,
   MLIRContext *ctx = builder.getContext();
   // TODO: Store clauses in op: allocateVars, allocatorVars, privateVars,
   // privatizers.
-  WsloopOp::build(
-      builder, state, clauses.linearVars, clauses.linearStepVars,
-      clauses.reductionVars, makeArrayAttr(ctx, clauses.reductionDeclSymbols),
-      clauses.scheduleValAttr, clauses.scheduleChunkVar,
-      clauses.scheduleModAttr, clauses.scheduleSimdAttr, clauses.nowaitAttr,
-      clauses.reductionByRefAttr, clauses.orderedAttr, clauses.orderAttr);
+  WsloopOp::build(builder, state, clauses.linearVars, clauses.linearStepVars,
+                  clauses.reductionVars,
+                  DenseBoolArrayAttr::get(ctx, clauses.reduceVarByRef),
+                  makeArrayAttr(ctx, clauses.reductionDeclSymbols),
+                  clauses.scheduleValAttr, clauses.scheduleChunkVar,
+                  clauses.scheduleModAttr, clauses.scheduleSimdAttr,
+                  clauses.nowaitAttr, clauses.orderedAttr, clauses.orderAttr);
 }
 
 LogicalResult WsloopOp::verify() {
@@ -1668,6 +1709,13 @@ LogicalResult WsloopOp::verify() {
       return emitError() << "only supported nested wrapper is 'omp.simd'";
   }
 
+  auto reductionVarsByRef = getReductionVarsByref();
+  if (reductionVarsByRef &&
+      reductionVarsByRef->size() != getReductionVars().size())
+    return emitOpError()
+           << "expected as many reduction variable by reference attributes "
+              "as reduction variables";
+
   return verifyReductionVarList(*this, getReductions(), getReductionVars());
 }
 
diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
index a7294632d6667..fa8c12dbca011 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
@@ -22,6 +22,7 @@
 #include "mlir/Target/LLVMIR/ModuleTranslation.h"
 #include "mlir/Transforms/RegionUtils.h"
 
+#include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/SetVector.h"
 #include "llvm/ADT/TypeSwitch.h"
 #include "llvm/Frontend/OpenMP/OMPConstants.h"
@@ -813,13 +814,16 @@ static void allocByValReductionVars(
     llvm::OpenMPIRBuilder::InsertPointTy &allocaIP,
     SmallVectorImpl<omp::DeclareReductionOp> &reductionDecls,
     SmallVectorImpl<llvm::Value *> &privateReductionVariables,
-    DenseMap<Value, llvm::Value *> &reductionVariableMap) {
+    DenseMap<Value, llvm::Value *> &reductionVariableMap,
+    llvm::ArrayRef<bool> isByRefs) {
   llvm::IRBuilderBase::InsertPointGuard guard(builder);
   builder.restoreIP(allocaIP);
   auto args =
       loop.getRegion().getArguments().take_back(loop.getNumReductionVars());
 
   for (std::size_t i = 0; i < loop.getNumReductionVars(); ++i) {
+    if (isByRefs[i])
+      continue;
     llvm::Value *var = builder.CreateAlloca(
         moduleTranslation.convertType(reductionDecls[i].getType()));
     moduleTranslation.mapValue(args[i], var);
@@ -919,13 +923,21 @@ inlineOmpRegionCleanup(llvm::SmallVectorImpl<Region *> &cleanupRegions,
   return success();
 }
 
+static ArrayRef<bool> getIsByRef(std::optional<ArrayRef<bool>> attr) {
+  if (!attr)
+    return {};
+  return *attr;
+}
+
 /// Converts an OpenMP workshare loop into LLVM IR using OpenMPIRBuilder.
 static LogicalResult
 convertOmpWsloop(Operation &opInst, llvm::IRBuilderBase &builder,
                  LLVM::ModuleTranslation &moduleTranslation) {
   auto wsloopOp = cast<omp::WsloopOp>(opInst);
   auto loopOp = cast<omp::LoopNestOp>(wsloopOp.getWrappedLoop());
-  const bool isByRef = wsloopOp.getByref();
+
+  llvm::ArrayRef<bool> isByRef = getIsByRef(wsloopOp.getReductionVarsByref());
+  assert(isByRef.size() == wsloopOp.getNumReductionVars());
 
   // TODO: this should be in the op verifier instead.
   if (loopOp.getLowerBound().empty())
@@ -952,11 +964,9 @@ convertOmpWsloop(Operation &opInst, llvm::IRBuilderBase &builder,
 
   SmallVector<llvm::Value *> privateReductionVariables;
   DenseMap<Value, llvm::Value *> reductionVariableMap;
-  if (!isByRef) {
-    allocByValReductionVars(wsloopOp, builder, moduleTranslation, allocaIP,
-                            reductionDecls, privateReductionVariables,
-                            reductionVariableMap);
-  }
+  allocByValReductionVars(wsloopOp, builder, moduleTranslation, allocaIP,
+                          reductionDecls, privateReductionVariables,
+                          reductionVariableMap, isByRef);
 
   // Before the loop, store the initial values of reductions into reduction
   // variables. Although this could be done after allocas, we don't want to mess
@@ -974,7 +984,7 @@ convertOmpWsloop(Operation &opInst, llvm::IRBuilderBase &builder,
       return failure();
     assert(phis.size() == 1 && "expected one value to be yielded from the "
                                "reduction neutral element declaration region");
-    if (isByRef) {
+    if (isByRef[i]) {
       // Allocate reduction variable (which is a pointer to the real reduction
       // variable allocated in the inlined region)
       llvm::Value *var = builder.CreateAlloca(
@@ -1108,7 +1118,7 @@ convertOmpWsloop(Operation &opInst, llvm::IRBuilderBase &builder,
   builder.SetInsertPoint(tempTerminator);
   llvm::OpenMPIRBuilder::InsertPointTy contInsertPoint =
       ompBuilder->createReductions(builder.saveIP(), allocaIP, reductionInfos,
-                                   wsloopOp.getNowait(), isByRef);
+                                   isByRef, wsloopOp.getNowait());
   if (!contInsertPoint.getBlock())
     return wsloopOp->emitOpError() << "failed to convert reductions";
   auto nextInsertionPoint =
@@ -1170,7 +1180,8 @@ convertOmpParallel(omp::ParallelOp opInst, llvm::IRBuilderBase &builder,
                    LLVM::ModuleTranslation &moduleTranslation) {
   using InsertPointTy = llvm::OpenMPIRBuilder::InsertPointTy;
   OmpParallelOpConversionManager raii(opInst);
-  const bool isByRef = opInst.getByref();
+  ArrayRef<bool> isByRef = getIsByRef(opInst.getReductionVarsByref());
+  assert(isByRef.size() == opInst.getNumReductionVars());
 
   // TODO: support error propagation in OpenMPIRBuilder and use it instead of
   // relying on captured variables.
@@ -1185,11 +1196,9 @@ convertOmpParallel(omp::ParallelOp opInst, llvm::IRBuilderBase &builder,
   auto bodyGenCB = [&](InsertPointTy allocaIP, InsertPointTy codeGenIP) {
     // Allocate reduction vars
     DenseMap<Value, llvm::Value *> reductionVariableMap;
-    if (!isByRef) {
-      allocByValReductionVars(opInst, builder, moduleTranslation, allocaIP,
-                              reductionDecls, privateReductionVariables,
-                              reductionVariableMap);
-    }
+    allocByValReductionVars(opInst, builder, moduleTranslation, allocaIP,
+                            reductionDecls, privateReductionVariables,
+                            reductionVariableMap, isByRef);
 
     // Initialize reduction vars
     builder.restoreIP(allocaIP);
@@ -1210,7 +1219,7 @@ convertOmpParallel(omp::ParallelOp opInst, llvm::IRBuilderBase &builder,
              "reduction neutral element declaration region");
       builder.restoreIP(allocaIP);
 
-      if (isByRef) {
+      if (isByRef[i]) {
         // Allocate reduction variable (which is a pointer to the real reduciton
         // variable allocated in the inlined region)
         llvm::Value *var = builder.CreateAlloca(
@@ -1269,7 +1278,7 @@ convertOmpParallel(omp::ParallelOp opInst, llvm::IRBuilderBase &builder,
 
       llvm::OpenMPIRBuilder::InsertPointTy contInsertPoint =
           ompBuilder->createReductions(builder.saveIP(), allocaIP,
-                                       reductionInfos, false, isByRef);
+                                       reductionInfos, isByRef, false);
       if (!contInsertPoint.getBlock()) {
         bodyGenStatus = opInst->emitOpError() << "failed to convert reductions";
         return;
diff --git a/mlir/test/Dialect/OpenMP/ops.mlir b/mlir/test/Dialect/OpenMP/ops.mlir
index 420cb226d593b..cfcb40d81bb34 100644
--- a/mlir/test/Dialect/OpenMP/ops.mlir
+++ b/mlir/test/Dialect/OpenMP/ops.mlir
@@ -869,6 +869,28 @@ func.func @wsloop_reduction(%lb : index, %ub : index, %step : index) {
   return
 }
 
+// CHECK-LABEL: func @wsloop_reduction_byref
+func.func @wsloop_reduction_byref(%lb : index, %ub : index, %step : index) {
+  %c1 = arith.constant 1 : i32
+  %0 = llvm.alloca %c1 x i32 : (i32) -> !llvm.ptr
+  // CHECK: reduction(byref @add_f32 %{{.+}} -> %[[PRV:.+]] : !llvm.ptr)
+  omp.wsloop reduction(byref @add_f32 %0 -> %prv : !llvm.ptr) {
+    omp.loop_nest (%iv) : index = (%lb) to (%ub) step (%step) {
+      // CHECK: %[[CST:.+]] = arith.constant 2.0{{.*}} : f32
+      %cst = arith.constant 2.0 : f32
+      // CHECK: %[[LPRV:.+]] = llvm.load %[[PRV]] : !llvm.ptr -> f32
+      %lprv = llvm.load %prv : !llvm.ptr -> f32
+      // CHECK: %[[RES:.+]] = llvm.fadd %[[LPRV]], %[[CST]] : f32
+      %res = llvm.fadd %lprv, %cst: f32
+      // CHECK: llvm.store %[[RES]], %[[PRV]] :  f32, !llvm.ptr
+      llvm.store %res, %prv :  f32, !llvm.ptr
+      omp.yield
+    }
+    omp.terminator
+  }
+  return
+}
+
 // CHECK-LABEL: func @parallel_reduction
 func.func @parallel_reduction() {
   %c1 = arith.constant 1 : i32
@@ -885,6 +907,22 @@ func.func @parallel_reduction() {
   return
 }
 
+// CHECK-LABEL: func @parallel_reduction_byref
+func.func @parallel_reduction_byref() {
+  %c1 = arith.constant 1 : i32
+  %0 = llvm.alloca %c1 x i32 : (i32) -> !llvm.ptr
+  // CHECK: omp.parallel reduction(byref @add_f32 {{.+}} -> {{.+}} : !llvm.ptr)
+  omp.parallel reduction(byref @add_f32 %0 -> %prv : !llvm.ptr) {
+    %1 = arith.constant 2.0 : f32
+    %2 = llvm.load %prv : !llvm.ptr -> f32
+    // CHECK: llvm.fadd %{{.*}}, %{{.*}} : f32
+    %3 = llvm.fadd %1, %2 : f32
+    llvm.store %3, %prv : f32, !llvm.ptr
+    omp.terminator
+  }
+  return
+}
+
 // CHECK: func @parallel_wsloop_reduction
 func.func @parallel_wsloop_reduction(%lb : index, %ub : index, %step : index) {
   %c1 = arith.constant 1 : i32
diff --git a/mlir/test/Target/LLVMIR/openmp-parallel-reduction-cleanup.mlir b/mlir/test/Target/LLVMIR/openmp-parallel-reduction-cleanup.mlir
index b7f71f438e56b..84a487cb8c98f 100644
--- a/mlir/test/Target/LLVMIR/openmp-parallel-reduction-cleanup.mlir
+++ b/mlir/test/Target/LLVMIR/openmp-parallel-reduction-cleanup.mlir
@@ -27,7 +27,7 @@
     %0 = llvm.mlir.constant(-1 : i32) : i32
     %1 = llvm.mlir.addressof @i : !llvm.ptr
     %2 = llvm.mlir.addressof @j : !llvm.ptr
-    omp.parallel byref reduction(@add_reduction_i_32 %1 -> %arg0 : !llvm.ptr, @add_reduction_i_32 %2 -> %arg1 : !llvm.ptr) {
+    omp.parallel reduction(byref @add_reduction_i_32 %1 -> %arg0 : !llvm.ptr, byref @add_reduction_i_32 %2 -> %arg1 : !llvm.ptr) {
       llvm.store %0, %arg0 : i32, !llvm.ptr
       llvm.store %0, %arg1 : i32, !llvm.ptr
       omp.terminator
diff --git a/mlir/test/Target/LLVMIR/openmp-reduction-byref.mlir b/mlir/test/Target/LLVMIR/openmp-reduction-byref.mlir
index e720969e82c11..ef1284547a88a 100644
--- a/mlir/test/Target/LLVMIR/openmp-reduction-byref.mlir
+++ b/mlir/test/Target/LLVMIR/openmp-reduction-byref.mlir
@@ -20,7 +20,7 @@
   llvm.func @main()  {
     %0 = llvm.mlir.constant(-1 : i32) : i32
     %1 = llvm.mlir.addressof @i : !llvm.ptr
-    omp.parallel byref reduction(@add_reduction_i_32 %1 -> %arg0 : !llvm.ptr) {
+    omp.parallel reduction(byref @add_reduction_i_32 %1 -> %arg0 : !llvm.ptr) {
       llvm.store %0, %arg0 : i32, !llvm.ptr
       omp.terminator
     }
diff --git a/mlir/test/Target/LLVMIR/openmp-reduction-init-arg.mlir b/mlir/test/Target/LLVMIR/openmp-reduction-init-arg.mlir
index 5dd31c425566c..f4b77cbf413d4 100644
--- a/mlir/test/Target/LLVMIR/openmp-reduction-init-arg.mlir
+++ b/mlir/test/Target/LLVMIR/openmp-reduction-init-arg.mlir
@@ -24,7 +24,7 @@ module {
   %87 = llvm.alloca %86 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> : (i64) -> !llvm.ptr
 // test multiple reduction variables to ensure they don't intefere with eachother
 // when inlining the reduction init region multiple times
-    omp.parallel byref reduction(@add_reduction_byref_box_Uxf64 %84 -> %arg3 : !llvm.ptr, @add_reduction_byref_box_Uxf64 %87 -> %arg4 : !llvm.ptr) {
+    omp.parallel reduction(byref @add_reduction_byref_box_Uxf64 %84 -> %arg3 : !llvm.ptr, byref @add_reduction_byref_box_Uxf64 %87 -> %arg4 : !llvm.ptr) {
       omp.terminator
     }
     llvm.return
diff --git a/mlir/test/Target/LLVMIR/openmp-wsloop-reduction-cleanup.mlir b/mlir/test/Target/LLVMIR/openmp-wsloop-reduction-cleanup.mlir
index 7a1a31830ce9b..0b28294dc1ced 100644
--- a/mlir/test/Target/LLVMIR/openmp-wsloop-reduction-cleanup.mlir
+++ b/mlir/test/Target/LLVMIR/openmp-wsloop-reduction-cleanup.mlir
@@ -30,7 +30,7 @@
     %loop_ub = llvm.mlir.constant(9 : i32) : i32
     %loop_lb = llvm.mlir.constant(0 : i32) : i32
     %loop_step = llvm.mlir.constant(1 : i32) : i32 
-    omp.wsloop byref reduction(@add_reduction_i_32 %1 -> %arg0 : !llvm.ptr, @add_reduction_i_32 %2 -> %arg1 : !llvm.ptr) {
+    omp.wsloop reduction(byref @add_reduction_i_32 %1 -> %arg0 : !llvm.ptr, byref @add_reduction_i_32 %2 -> %arg1 : !llvm.ptr) {
       omp.loop_nest (%loop_cnt) : i32 = (%loop_lb) to (%loop_ub) inclusive step (%loop_step) {
         llvm.store %0, %arg0 : i32, !llvm.ptr
         llvm.store %0, %arg1 : i32, !llvm.ptr



More information about the Mlir-commits mailing list