[flang-commits] [flang] cdf59b3 - [Flang][OpenMP] Add lowering and translation for the threadset clause (#215513)
via flang-commits
flang-commits at lists.llvm.org
Wed Sep 9 20:51:58 PDT 2026
Author: Ritanya-B-Bharadwaj
Date: 2026-09-10T09:21:51+05:30
New Revision: cdf59b384a48610a6450011279005b1e34f8572d
URL: https://github.com/llvm/llvm-project/commit/cdf59b384a48610a6450011279005b1e34f8572d
DIFF: https://github.com/llvm/llvm-project/commit/cdf59b384a48610a6450011279005b1e34f8572d.diff
LOG: [Flang][OpenMP] Add lowering and translation for the threadset clause (#215513)
Adds Flang lowering, and LLVM IR translation for the OpenMP 6.0 `threadset` clause (spec 14.8) on `task` and `taskloop`.
- Add a `ThreadsetPolicy` enum (`omp_pool`/`omp_team`) and clause on `omp.task` and `omp.taskloop.context`, lowered from Flang.
- Translate `threadset(omp_pool)` to the free-agent task flag (`0x80`); `omp_team` leaves it unset, matching clang.
- Add semantics, lowering, and translation tests.
Related: https://github.com/llvm/llvm-project/pull/135807,
https://github.com/llvm/llvm-project/pull/144409,
https://github.com/llvm/llvm-project/pull/169856
Added:
flang/test/Lower/OpenMP/threadset.f90
flang/test/Semantics/OpenMP/threadset-clause-v60.f90
mlir/test/Target/LLVMIR/openmp-task-threadset.mlir
mlir/test/Target/LLVMIR/openmp-taskloop-threadset.mlir
Modified:
flang/docs/OpenMPSupport.md
flang/lib/Lower/OpenMP/ClauseProcessor.cpp
flang/lib/Lower/OpenMP/ClauseProcessor.h
flang/lib/Lower/OpenMP/OpenMP.cpp
flang/test/Semantics/OpenMP/threadset-clause.f90
llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td
mlir/include/mlir/Dialect/OpenMP/OpenMPEnums.td
mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
mlir/test/Dialect/OpenMP/ops.mlir
Removed:
flang/test/Lower/OpenMP/Todo/threadset.f90
################################################################################
diff --git a/flang/docs/OpenMPSupport.md b/flang/docs/OpenMPSupport.md
index 5a5285421e724..ed14e6a2ab99f 100644
--- a/flang/docs/OpenMPSupport.md
+++ b/flang/docs/OpenMPSupport.md
@@ -245,7 +245,7 @@ Parser/Semantics, MLIR, Lowering, or the OpenMPIRBuilder.
| Feature | Status | Claimed By | Notes | Reviews |
|:--------|:-------|:-----------|:------|:--------|
-| threadset clause | <span class="part">partial</span> | | Semantics coverage exists (`flang/test/Semantics/OpenMP/threadset-clause.f90`) with ongoing lowering/runtime validation. | [llvm/llvm-project#169856](https://github.com/llvm/llvm-project/pull/169856) |
+| threadset clause | <span class="part">partial</span> | | Frontend (parsing + semantics) and lowering coverage exists (`flang/test/Semantics/OpenMP/threadset-clause.f90`, `flang/test/Semantics/OpenMP/threadset-clause-v60.f90`, `flang/test/Lower/OpenMP/threadset.f90`); the clause is represented on `omp.task`/`omp.taskloop.context` and translated to the runtime free-agent task flag. | [llvm/llvm-project#169856](https://github.com/llvm/llvm-project/pull/169856) |
| groupprivate directive | <span class="part">partial</span> | | Semantics and lowering coverage exists (`flang/test/Semantics/OpenMP/groupprivate.f90`, `flang/test/Lower/OpenMP/groupprivate.f90`, `flang/test/Lower/OpenMP/groupprivate-modfile.f90`); lowering in non-teams contexts not fully handled. | [llvm/llvm-project#166199](https://github.com/llvm/llvm-project/pull/166199), [llvm/llvm-project#166214](https://github.com/llvm/llvm-project/pull/166214), [llvm/llvm-project#180934](https://github.com/llvm/llvm-project/pull/180934) |
| recording of task graphs | <span class="progress">in progress</span> | | Semantics coverage exists (`flang/test/Semantics/OpenMP/taskgraph.f90`). | |
| workdistribute construct | <span class="part">partial</span> | | Semantics/lowering/transform coverage exists (`flang/test/Semantics/OpenMP/workdistribute01.f90`, `flang/test/Lower/OpenMP/workdistribute.f90`, `flang/test/Transforms/OpenMP/lower-workdistribute-fission.mlir`) including `target teams` placement updates; some team-nesting combinations still intentionally diagnose as unsupported. | [llvm/llvm-project#154377](https://github.com/llvm/llvm-project/pull/154377), [llvm/llvm-project#154378](https://github.com/llvm/llvm-project/pull/154378), [llvm/llvm-project#140523](https://github.com/llvm/llvm-project/pull/140523), [llvm/llvm-project#199006](https://github.com/llvm/llvm-project/pull/199006) |
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
index 22506b14b0fa6..b871f630d3dae 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.cpp
@@ -985,6 +985,22 @@ bool ClauseProcessor::processThreadLimit(
return false;
}
+bool ClauseProcessor::processThreadset(
+ mlir::omp::ThreadsetClauseOps &result) const {
+ using Threadset = omp::clause::Threadset;
+ if (auto *clause = findUniqueClause<Threadset>()) {
+ fir::FirOpBuilder &firOpBuilder = converter.getFirOpBuilder();
+ mlir::omp::ThreadsetPolicy policy =
+ clause->v == Threadset::ThreadsetPolicy::Omp_Pool
+ ? mlir::omp::ThreadsetPolicy::omp_pool
+ : mlir::omp::ThreadsetPolicy::omp_team;
+ result.threadset =
+ mlir::omp::ThreadsetPolicyAttr::get(firOpBuilder.getContext(), policy);
+ return true;
+ }
+ return false;
+}
+
bool ClauseProcessor::processUntied(mlir::omp::UntiedClauseOps &result) const {
return markClauseOccurrence<omp::clause::Untied>(result.untied);
}
diff --git a/flang/lib/Lower/OpenMP/ClauseProcessor.h b/flang/lib/Lower/OpenMP/ClauseProcessor.h
index 6bfa939f52c0b..6b0e0cf606f48 100644
--- a/flang/lib/Lower/OpenMP/ClauseProcessor.h
+++ b/flang/lib/Lower/OpenMP/ClauseProcessor.h
@@ -125,6 +125,7 @@ class ClauseProcessor {
bool processSimd(mlir::omp::OrderedRegionOperands &result) const;
bool processThreadLimit(lower::StatementContext &stmtCtx,
mlir::omp::ThreadLimitClauseOps &result) const;
+ bool processThreadset(mlir::omp::ThreadsetClauseOps &result) const;
bool processUntied(mlir::omp::UntiedClauseOps &result) const;
bool processDetach(mlir::omp::DetachClauseOps &result) const;
// 'Repeatable' clauses: They can appear multiple times in the clause list.
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index 8ea1041adcbbf..89f768c6dc811 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -2746,6 +2746,7 @@ static void genTaskClauses(lower::AbstractConverter &converter,
cp.processInReduction(loc, clauseOps, inReductionObjects);
cp.processMergeable(clauseOps);
cp.processPriority(stmtCtx, clauseOps);
+ cp.processThreadset(clauseOps);
cp.processUntied(clauseOps);
cp.processDetach(clauseOps);
}
@@ -2779,6 +2780,7 @@ static void genTaskloopClauses(
cp.processNumTasks(stmtCtx, clauseOps);
cp.processPriority(stmtCtx, clauseOps);
cp.processReduction(loc, clauseOps, reductionObjects);
+ cp.processThreadset(clauseOps);
cp.processUntied(clauseOps);
}
@@ -8196,6 +8198,7 @@ static void genOMP(lower::AbstractConverter &converter, lower::SymMap &symTable,
!std::holds_alternative<clause::Simd>(clause.u) &&
!std::holds_alternative<clause::ThreadLimit>(clause.u) &&
!std::holds_alternative<clause::Threads>(clause.u) &&
+ !std::holds_alternative<clause::Threadset>(clause.u) &&
!std::holds_alternative<clause::UseDeviceAddr>(clause.u) &&
!std::holds_alternative<clause::UseDevicePtr>(clause.u) &&
!std::holds_alternative<clause::InReduction>(clause.u) &&
diff --git a/flang/test/Lower/OpenMP/Todo/threadset.f90 b/flang/test/Lower/OpenMP/Todo/threadset.f90
deleted file mode 100644
index b022baf02654b..0000000000000
--- a/flang/test/Lower/OpenMP/Todo/threadset.f90
+++ /dev/null
@@ -1,10 +0,0 @@
-! RUN: %not_todo_cmd %flang_fc1 -emit-fir -fopenmp -fopenmp-version=60 -o - %s 2>&1 | FileCheck %s
-
-! CHECK: not yet implemented: THREADSET clause is not implemented yet
-
-subroutine f00(x)
- integer :: x(10)
- !$omp task threadset(omp_pool)
- x = x + 1
- !$omp end task
-end
diff --git a/flang/test/Lower/OpenMP/threadset.f90 b/flang/test/Lower/OpenMP/threadset.f90
new file mode 100644
index 0000000000000..c758e160afc01
--- /dev/null
+++ b/flang/test/Lower/OpenMP/threadset.f90
@@ -0,0 +1,37 @@
+!RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-version=60 %s -o - | FileCheck %s
+
+!CHECK-LABEL: func @_QPtask_threadset_pool
+subroutine task_threadset_pool()
+ !CHECK: omp.task threadset(omp_pool) {
+ !$omp task threadset(omp_pool)
+ !CHECK: omp.terminator
+ !$omp end task
+end subroutine task_threadset_pool
+
+!CHECK-LABEL: func @_QPtask_threadset_team
+subroutine task_threadset_team()
+ !CHECK: omp.task threadset(omp_team) {
+ !$omp task threadset(omp_team)
+ !CHECK: omp.terminator
+ !$omp end task
+end subroutine task_threadset_team
+
+!CHECK-LABEL: func @_QPtaskloop_threadset_pool
+subroutine taskloop_threadset_pool()
+ integer :: i
+ !CHECK: omp.taskloop.context threadset(omp_pool)
+ !$omp taskloop threadset(omp_pool)
+ do i = 1, 10
+ end do
+ !$omp end taskloop
+end subroutine taskloop_threadset_pool
+
+!CHECK-LABEL: func @_QPtaskloop_threadset_team
+subroutine taskloop_threadset_team()
+ integer :: i
+ !CHECK: omp.taskloop.context threadset(omp_team)
+ !$omp taskloop threadset(omp_team)
+ do i = 1, 10
+ end do
+ !$omp end taskloop
+end subroutine taskloop_threadset_team
diff --git a/flang/test/Semantics/OpenMP/threadset-clause-v60.f90 b/flang/test/Semantics/OpenMP/threadset-clause-v60.f90
new file mode 100644
index 0000000000000..f5542e4941cc2
--- /dev/null
+++ b/flang/test/Semantics/OpenMP/threadset-clause-v60.f90
@@ -0,0 +1,38 @@
+!RUN: %python %S/../test_errors.py %s %flang -fopenmp -fopenmp-version=60
+
+subroutine f00(x)
+ integer :: x(10)
+ integer :: i
+
+! Valid uses on TASK.
+!$omp task threadset(omp_pool)
+ x = x + 1
+!$omp end task
+
+!$omp task threadset(omp_team)
+ x = x + 1
+!$omp end task
+
+! Valid uses on TASKLOOP.
+!$omp taskloop threadset(omp_pool)
+ do i = 1, 10
+ end do
+!$omp end taskloop
+
+!$omp taskloop threadset(omp_team)
+ do i = 1, 10
+ end do
+!$omp end taskloop
+
+! At most one THREADSET clause is allowed on the directive.
+!ERROR: At most one THREADSET clause can appear on TASK directive
+!$omp task threadset(omp_pool) threadset(omp_team)
+ x = x + 1
+!$omp end task
+
+! THREADSET is only allowed on TASK and TASKLOOP.
+!ERROR: THREADSET clause is not allowed on PARALLEL directive
+!$omp parallel threadset(omp_pool)
+ x = x + 1
+!$omp end parallel
+end
diff --git a/flang/test/Semantics/OpenMP/threadset-clause.f90 b/flang/test/Semantics/OpenMP/threadset-clause.f90
index 9c811d440fb6a..65a7ed74037cc 100644
--- a/flang/test/Semantics/OpenMP/threadset-clause.f90
+++ b/flang/test/Semantics/OpenMP/threadset-clause.f90
@@ -2,8 +2,15 @@
subroutine f00(x)
integer :: x(10)
+ integer :: i
!ERROR: THREADSET clause is not allowed on TASK directive in OpenMP v4.5, try -fopenmp-version=60
!$omp task threadset(omp_pool)
x = x + 1
!$omp end task
+
+!ERROR: THREADSET clause is not allowed on TASKLOOP directive in OpenMP v4.5, try -fopenmp-version=60
+!$omp taskloop threadset(omp_team)
+ do i = 1, 10
+ end do
+!$omp end taskloop
end
diff --git a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
index 5f3da79cf53b1..e4fffae9a7bbe 100644
--- a/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
+++ b/llvm/include/llvm/Frontend/OpenMP/OMPIRBuilder.h
@@ -1628,6 +1628,8 @@ class OpenMPIRBuilder {
/// \param TaskContextStructPtrVal If non-null, a pointer to to be placed
/// immediately after the {lower bound, upper
/// bound, step} values in the task data.
+ /// \param FreeAgent If `true`, the generated tasks are eligible to be
+ /// executed by a free-agent thread (threadset(omp_pool)).
LLVM_ABI InsertPointOrErrorTy createTaskloop(
const LocationDescription &Loc, InsertPointTy AllocaIP,
ArrayRef<BasicBlock *> DeallocBlocks, BodyGenCallbackTy BodyGenCB,
@@ -1637,7 +1639,7 @@ class OpenMPIRBuilder {
int Sched = 0, Value *Final = nullptr, bool Mergeable = false,
Value *Priority = nullptr, uint64_t NumOfCollapseLoops = 1,
TaskDupCallbackTy DupCB = nullptr,
- Value *TaskContextStructPtrVal = nullptr);
+ Value *TaskContextStructPtrVal = nullptr, bool FreeAgent = false);
/// Generator for `#omp task`
///
@@ -1664,13 +1666,16 @@ class OpenMPIRBuilder {
/// \param Mergeable If the given task is `mergeable`
/// \param priority `priority-value' specifies the execution order of the
/// tasks that is generated by the construct
+ /// \param FreeAgent If `true`, the task is eligible to be executed by a
+ /// free-agent thread (threadset(omp_pool)).
LLVM_ABI InsertPointOrErrorTy createTask(
const LocationDescription &Loc, InsertPointTy AllocaIP,
ArrayRef<BasicBlock *> DeallocBlocks, BodyGenCallbackTy BodyGenCB,
bool Tied = true, Value *Final = nullptr, Value *IfCondition = nullptr,
const DependenciesInfo &Dependencies = {},
const AffinityData &Affinities = {}, bool Mergeable = false,
- Value *EventHandle = nullptr, Value *Priority = nullptr);
+ Value *EventHandle = nullptr, Value *Priority = nullptr,
+ bool FreeAgent = false);
/// Generator for the taskgroup construct
///
diff --git a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
index 8d6a35a368b4f..1a9ed6d9f69ea 100644
--- a/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
+++ b/llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp
@@ -2549,7 +2549,7 @@ OpenMPIRBuilder::InsertPointOrErrorTy OpenMPIRBuilder::createTaskloop(
Value *LBVal, Value *UBVal, Value *StepVal, bool Untied, Value *IfCond,
Value *GrainSize, bool NoGroup, int Sched, Value *Final, bool Mergeable,
Value *Priority, uint64_t NumOfCollapseLoops, TaskDupCallbackTy DupCB,
- Value *TaskContextStructPtrVal) {
+ Value *TaskContextStructPtrVal, bool FreeAgent) {
if (!updateToLocation(Loc))
return InsertPointTy();
@@ -2629,7 +2629,8 @@ OpenMPIRBuilder::InsertPointOrErrorTy OpenMPIRBuilder::createTaskloop(
TaskloopAllocaBB, CLI, TaskDupFn, ToBeDeleted, IfCond,
GrainSize, NoGroup, Sched, FakeLB, FakeUB, FakeStep,
FakeSharedsTy, Final, Mergeable, Priority,
- NumOfCollapseLoops](Function &OutlinedFn) mutable {
+ NumOfCollapseLoops,
+ FreeAgent](Function &OutlinedFn) mutable {
// Replace the Stale CI by appropriate RTL function call.
assert(OutlinedFn.hasOneUse() &&
"there must be a single user for the outlined function");
@@ -2671,6 +2672,8 @@ OpenMPIRBuilder::InsertPointOrErrorTy OpenMPIRBuilder::createTaskloop(
// Task is not mergeable if (Flags & 4) == 0.
// Task is priority if (Flags & 32) == 32.
// Task is not priority if (Flags & 32) == 0.
+ // Task is free-agent eligible if (Flags & 128) == 128.
+ // Task is not free-agent eligible if (Flags & 128) == 0.
Value *Flags = Builder.getInt32(Untied ? 0 : 1);
if (Final)
Flags = Builder.CreateOr(Builder.getInt32(2), Flags);
@@ -2678,6 +2681,8 @@ OpenMPIRBuilder::InsertPointOrErrorTy OpenMPIRBuilder::createTaskloop(
Flags = Builder.CreateOr(Builder.getInt32(4), Flags);
if (Priority)
Flags = Builder.CreateOr(Builder.getInt32(32), Flags);
+ if (FreeAgent)
+ Flags = Builder.CreateOr(Builder.getInt32(128), Flags);
Value *TaskSize = Builder.getInt64(
divideCeil(M.getDataLayout().getTypeSizeInBits(Task), 8));
@@ -2897,7 +2902,7 @@ OpenMPIRBuilder::InsertPointOrErrorTy OpenMPIRBuilder::createTask(
ArrayRef<BasicBlock *> DeallocBlocks, BodyGenCallbackTy BodyGenCB,
bool Tied, Value *Final, Value *IfCondition,
const DependenciesInfo &Dependencies, const AffinityData &Affinities,
- bool Mergeable, Value *EventHandle, Value *Priority) {
+ bool Mergeable, Value *EventHandle, Value *Priority, bool FreeAgent) {
if (!updateToLocation(Loc))
return InsertPointTy();
@@ -2945,7 +2950,7 @@ OpenMPIRBuilder::InsertPointOrErrorTy OpenMPIRBuilder::createTask(
Builder, AllocaIP, ToBeDeleted, TaskAllocaIP, "global.tid", false));
OI->PostOutlineCB = [this, Ident, Tied, Final, IfCondition, Dependencies,
- Affinities, Mergeable, Priority, EventHandle,
+ Affinities, Mergeable, Priority, EventHandle, FreeAgent,
TaskAllocaBB,
ToBeDeleted](Function &OutlinedFn) mutable {
// Replace the Stale CI by appropriate RTL function call.
@@ -2978,6 +2983,8 @@ OpenMPIRBuilder::InsertPointOrErrorTy OpenMPIRBuilder::createTask(
// Task is not detachable iff (Flags & 64) == 0.
// Task is priority iff (Flags & 32) == 32.
// Task is not priority iff (Flags & 32) == 0.
+ // Task is free-agent eligible iff (Flags & 128) == 128.
+ // Task is not free-agent eligible iff (Flags & 128) == 0.
// TODO: Handle the other flags.
Value *Flags = Builder.getInt32(Tied);
auto *ConstIfCondition = dyn_cast_or_null<ConstantInt>(IfCondition);
@@ -2994,6 +3001,8 @@ OpenMPIRBuilder::InsertPointOrErrorTy OpenMPIRBuilder::createTask(
Flags = Builder.CreateOr(Builder.getInt32(64), Flags);
if (Priority)
Flags = Builder.CreateOr(Builder.getInt32(32), Flags);
+ if (FreeAgent)
+ Flags = Builder.CreateOr(Builder.getInt32(128), Flags);
// Argument - `sizeof_kmp_task_t` (TaskSize)
// Tasksize refers to the size in bytes of kmp_task_t data structure
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td
index bd9f1321f077b..55a8bba40404b 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPClauses.td
@@ -1716,6 +1716,33 @@ class OpenMP_ThreadLimitClauseSkip<
def OpenMP_ThreadLimitClause : OpenMP_ThreadLimitClauseSkip<>;
+//===----------------------------------------------------------------------===//
+// V6.0: [14.8] `threadset` clause
+//===----------------------------------------------------------------------===//
+
+class OpenMP_ThreadsetClauseSkip<
+ bit traits = false, bit arguments = false, bit assemblyFormat = false,
+ bit description = false, bit extraClassDeclaration = false
+ > : OpenMP_Clause<traits, arguments, assemblyFormat, description,
+ extraClassDeclaration> {
+ let arguments = (ins
+ OptionalAttr<ThreadsetPolicyAttr>:$threadset
+ );
+
+ let optAssemblyFormat = [{
+ `threadset` `(` custom<ClauseAttr>($threadset) `)`
+ }];
+
+ let description = [{
+ The `threadset` clause specifies the set of threads that may execute the
+ generated task. When the policy is `omp_pool`, the task is eligible to be
+ executed by a free-agent thread. When the policy is `omp_team`, the task
+ may only be executed by a thread in the current team.
+ }];
+}
+
+def OpenMP_ThreadsetClause : OpenMP_ThreadsetClauseSkip<>;
+
//===----------------------------------------------------------------------===//
// V5.2: [9.1.1] `sizes` clause
//===----------------------------------------------------------------------===//
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPEnums.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPEnums.td
index f1b9c033ea0b9..50b02c0d9e61c 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPEnums.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPEnums.td
@@ -376,6 +376,20 @@ def FallbackModifierAttr : OpenMP_EnumAttr<FallbackModifier,
"fallback_modifier">;
//===----------------------------------------------------------------------===//
+// threadset_policy enum.
+//===----------------------------------------------------------------------===//
+
+def ThreadsetOmpPool : I32EnumAttrCase<"omp_pool", 0>;
+def ThreadsetOmpTeam : I32EnumAttrCase<"omp_team", 1>;
+
+def ThreadsetPolicy : OpenMP_I32EnumAttr<
+ "ThreadsetPolicy",
+ "threadset policy", [
+ ThreadsetOmpPool,
+ ThreadsetOmpTeam
+ ]>;
+
+def ThreadsetPolicyAttr : OpenMP_EnumAttr<ThreadsetPolicy, "threadset_policy">;
// interop_type for interop construct
//===----------------------------------------------------------------------===//
diff --git a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
index 0df33d96319f5..b2ad6abd40ed3 100644
--- a/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
+++ b/mlir/include/mlir/Dialect/OpenMP/OpenMPOps.td
@@ -1037,7 +1037,7 @@ def TaskOp : OpenMP_Op<"task", traits = [
OpenMP_FinalClause, OpenMP_IfClause,
OpenMP_InReductionClauseSkip<assemblyFormat = true>,
OpenMP_MergeableClause, OpenMP_PriorityClause, OpenMP_PrivateClause,
- OpenMP_UntiedClause, OpenMP_DetachClause
+ OpenMP_ThreadsetClause, OpenMP_UntiedClause, OpenMP_DetachClause
], singleRegion = true> {
let summary = "task construct";
let description = [{
@@ -1084,7 +1084,7 @@ def TaskloopContextOp : OpenMP_Op<"taskloop.context", traits = [
OpenMP_IfClause, OpenMP_InReductionClauseSkip<assemblyFormat = true>,
OpenMP_MergeableClause, OpenMP_NogroupClause, OpenMP_NumTasksClause,
OpenMP_PriorityClause, OpenMP_PrivateClause, OpenMP_ReductionClause,
- OpenMP_UntiedClause
+ OpenMP_ThreadsetClause, OpenMP_UntiedClause
], singleRegion = true> {
let summary = "OutlinableOpenMPOpInterface wrapper for taskloop construct";
let description = [{
diff --git a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
index 1f210ef60fe39..a74d0eb51564c 100644
--- a/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
+++ b/mlir/lib/Dialect/OpenMP/IR/OpenMPDialect.cpp
@@ -3843,19 +3843,20 @@ LogicalResult DeclareReductionOp::verifyRegions() {
void TaskOp::build(OpBuilder &builder, OperationState &state,
const TaskOperands &clauses) {
MLIRContext *ctx = builder.getContext();
- TaskOp::build(
- builder, state, clauses.iterated, clauses.affinityVars,
- clauses.allocateVars, clauses.allocatorVars,
- makeDenseI64ArrayAttr(ctx, clauses.allocateAlignments),
- makeDenseI64ArrayAttr(ctx, clauses.allocatePrivateIndices),
- makeArrayAttr(ctx, clauses.dependKinds), clauses.dependVars,
- makeArrayAttr(ctx, clauses.dependIteratedKinds), clauses.dependIterated,
- clauses.final, clauses.ifExpr, clauses.inReductionVars,
- makeDenseBoolArrayAttr(ctx, clauses.inReductionByref),
- makeArrayAttr(ctx, clauses.inReductionSyms), clauses.mergeable,
- clauses.priority, /*private_vars=*/clauses.privateVars,
- /*private_syms=*/makeArrayAttr(ctx, clauses.privateSyms),
- clauses.privateNeedsBarrier, clauses.untied, clauses.eventHandle);
+ TaskOp::build(builder, state, clauses.iterated, clauses.affinityVars,
+ clauses.allocateVars, clauses.allocatorVars,
+ makeDenseI64ArrayAttr(ctx, clauses.allocateAlignments),
+ makeDenseI64ArrayAttr(ctx, clauses.allocatePrivateIndices),
+ makeArrayAttr(ctx, clauses.dependKinds), clauses.dependVars,
+ makeArrayAttr(ctx, clauses.dependIteratedKinds),
+ clauses.dependIterated, clauses.final, clauses.ifExpr,
+ clauses.inReductionVars,
+ makeDenseBoolArrayAttr(ctx, clauses.inReductionByref),
+ makeArrayAttr(ctx, clauses.inReductionSyms), clauses.mergeable,
+ clauses.priority, /*private_vars=*/clauses.privateVars,
+ /*private_syms=*/makeArrayAttr(ctx, clauses.privateSyms),
+ clauses.privateNeedsBarrier, clauses.threadset, clauses.untied,
+ clauses.eventHandle);
}
LogicalResult TaskOp::verify() {
@@ -3925,7 +3926,8 @@ void TaskloopContextOp::build(OpBuilder &builder, OperationState &state,
/*private_syms=*/makeArrayAttr(ctx, clauses.privateSyms),
clauses.privateNeedsBarrier, clauses.reductionMod, clauses.reductionVars,
makeDenseBoolArrayAttr(ctx, clauses.reductionByref),
- makeArrayAttr(ctx, clauses.reductionSyms), clauses.untied);
+ makeArrayAttr(ctx, clauses.reductionSyms), clauses.threadset,
+ clauses.untied);
state.addAttribute("omp.combined", UnitAttr::get(ctx));
}
diff --git a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
index e6e784f37d6b2..1d3827783b31f 100644
--- a/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
+++ b/mlir/lib/Target/LLVMIR/Dialect/OpenMP/OpenMPToLLVMIRTranslation.cpp
@@ -3521,7 +3521,8 @@ convertOmpTaskOp(omp::TaskOp taskOp, llvm::IRBuilderBase &builder,
moduleTranslation.lookupValue(taskOp.getIfExpr()), dependencies, ad,
taskOp.getMergeable(),
moduleTranslation.lookupValue(taskOp.getEventHandle()),
- moduleTranslation.lookupValue(taskOp.getPriority()));
+ moduleTranslation.lookupValue(taskOp.getPriority()),
+ taskOp.getThreadset() == omp::ThreadsetPolicy::omp_pool);
if (failed(handleError(afterIP, *taskOp)))
return failure();
@@ -4158,7 +4159,8 @@ convertOmpTaskloopContextOp(omp::TaskloopContextOp contextOp,
contextOp.getMergeable(),
moduleTranslation.lookupValue(contextOp.getPriority()),
loopOp.getCollapseNumLoops(), taskDupOrNull,
- taskStructMgr.getStructPtr());
+ taskStructMgr.getStructPtr(),
+ contextOp.getThreadset() == omp::ThreadsetPolicy::omp_pool);
if (failed(handleError(afterIP, opInst)))
return failure();
diff --git a/mlir/test/Dialect/OpenMP/ops.mlir b/mlir/test/Dialect/OpenMP/ops.mlir
index 92cc73f59ae58..d20ed5871c8d1 100644
--- a/mlir/test/Dialect/OpenMP/ops.mlir
+++ b/mlir/test/Dialect/OpenMP/ops.mlir
@@ -2486,6 +2486,23 @@ func.func @omp_task(%bool_var: i1, %i64_var: i64, %i32_var: i32, %data_var: memr
omp.terminator
}
+ // Checking `threadset` clause
+ // CHECK: omp.task threadset(omp_pool) {
+ omp.task threadset(omp_pool) {
+ // CHECK: "test.foo"() : () -> ()
+ "test.foo"() : () -> ()
+ // CHECK: omp.terminator
+ omp.terminator
+ }
+
+ // CHECK: omp.task threadset(omp_team) {
+ omp.task threadset(omp_team) {
+ // CHECK: "test.foo"() : () -> ()
+ "test.foo"() : () -> ()
+ // CHECK: omp.terminator
+ omp.terminator
+ }
+
// Checking `in_reduction` clause
%c1 = arith.constant 1 : i32
// CHECK: %[[redn_var1:.*]] = llvm.alloca %{{.*}} x f32 : (i32) -> !llvm.ptr
@@ -3015,6 +3032,30 @@ func.func @omp_taskloop(%lb: i32, %ub: i32, %step: i32) -> () {
omp.terminator
} {omp.combined}
+ // CHECK: omp.taskloop.context threadset(omp_pool) {
+ omp.taskloop.context threadset(omp_pool) {
+ // CHECK: omp.taskloop.wrapper {
+ omp.taskloop.wrapper {
+ omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
+ // CHECK: omp.yield
+ omp.yield
+ }
+ }
+ omp.terminator
+ } {omp.combined}
+
+ // CHECK: omp.taskloop.context threadset(omp_team) {
+ omp.taskloop.context threadset(omp_team) {
+ // CHECK: omp.taskloop.wrapper {
+ omp.taskloop.wrapper {
+ omp.loop_nest (%i, %j) : i32 = (%lb, %ub) to (%ub, %lb) step (%step, %step) {
+ // CHECK: omp.yield
+ omp.yield
+ }
+ }
+ omp.terminator
+ } {omp.combined}
+
%testf32 = "test.f32"() : () -> (!llvm.ptr)
%testf32_2 = "test.f32"() : () -> (!llvm.ptr)
// CHECK: omp.taskloop.context in_reduction(@add_f32 %{{.+}} -> %{{.+}}, @add_f32 %{{.+}} -> %{{.+}} : !llvm.ptr, !llvm.ptr) {
diff --git a/mlir/test/Target/LLVMIR/openmp-task-threadset.mlir b/mlir/test/Target/LLVMIR/openmp-task-threadset.mlir
new file mode 100644
index 0000000000000..26aa1fb8ae0f2
--- /dev/null
+++ b/mlir/test/Target/LLVMIR/openmp-task-threadset.mlir
@@ -0,0 +1,26 @@
+// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s
+
+// Check that threadset(omp_pool) sets the free-agent task flag (0x80 = 128) and
+// threadset(omp_team) does not.
+
+llvm.func @task_threadset_pool() {
+ omp.task threadset(omp_pool) {
+ omp.terminator
+ }
+ llvm.return
+}
+
+// CHECK-LABEL: define void @task_threadset_pool()
+// CHECK: call ptr @__kmpc_omp_task_alloc(ptr @{{.+}}, i32 %{{.+}}, i32 129, i64 {{.+}}, i64 {{.+}}, ptr @{{.+}})
+
+// -----
+
+llvm.func @task_threadset_team() {
+ omp.task threadset(omp_team) {
+ omp.terminator
+ }
+ llvm.return
+}
+
+// CHECK-LABEL: define void @task_threadset_team()
+// CHECK: call ptr @__kmpc_omp_task_alloc(ptr @{{.+}}, i32 %{{.+}}, i32 1, i64 {{.+}}, i64 {{.+}}, ptr @{{.+}})
diff --git a/mlir/test/Target/LLVMIR/openmp-taskloop-threadset.mlir b/mlir/test/Target/LLVMIR/openmp-taskloop-threadset.mlir
new file mode 100644
index 0000000000000..332fee75ecdb6
--- /dev/null
+++ b/mlir/test/Target/LLVMIR/openmp-taskloop-threadset.mlir
@@ -0,0 +1,50 @@
+// RUN: mlir-translate -mlir-to-llvmir %s | FileCheck %s
+
+// Check that threadset(omp_pool) sets the free-agent task flag (0x80 = 128) and
+// threadset(omp_team) does not.
+
+omp.private {type = private} @_QFtestEi_private_i32 : i32
+
+llvm.func @_QPtest_pool() {
+ %0 = llvm.mlir.constant(1 : i64) : i64
+ %1 = llvm.alloca %0 x i32 {bindc_name = "i"} : (i64) -> !llvm.ptr
+ %7 = llvm.mlir.constant(1 : i32) : i32
+ %8 = llvm.mlir.constant(5 : i32) : i32
+ %9 = llvm.mlir.constant(1 : i32) : i32
+ omp.taskloop.context threadset(omp_pool) private(@_QFtestEi_private_i32 %1 -> %arg1 : !llvm.ptr) {
+ omp.taskloop.wrapper {
+ omp.loop_nest (%arg2) : i32 = (%7) to (%8) inclusive step (%9) {
+ llvm.store %arg2, %arg1 : i32, !llvm.ptr
+ omp.yield
+ }
+ }
+ omp.terminator
+ } {omp.combined}
+ llvm.return
+}
+
+// CHECK-LABEL: define void @_QPtest_pool()
+// CHECK: call ptr @__kmpc_omp_task_alloc(ptr @{{.+}}, i32 %{{.+}}, i32 129, i64 {{.+}}, i64 {{.+}}, ptr @_QPtest_pool..omp_par)
+
+// -----
+
+llvm.func @_QPtest_team() {
+ %0 = llvm.mlir.constant(1 : i64) : i64
+ %1 = llvm.alloca %0 x i32 {bindc_name = "i"} : (i64) -> !llvm.ptr
+ %7 = llvm.mlir.constant(1 : i32) : i32
+ %8 = llvm.mlir.constant(5 : i32) : i32
+ %9 = llvm.mlir.constant(1 : i32) : i32
+ omp.taskloop.context threadset(omp_team) private(@_QFtestEi_private_i32 %1 -> %arg1 : !llvm.ptr) {
+ omp.taskloop.wrapper {
+ omp.loop_nest (%arg2) : i32 = (%7) to (%8) inclusive step (%9) {
+ llvm.store %arg2, %arg1 : i32, !llvm.ptr
+ omp.yield
+ }
+ }
+ omp.terminator
+ } {omp.combined}
+ llvm.return
+}
+
+// CHECK-LABEL: define void @_QPtest_team()
+// CHECK: call ptr @__kmpc_omp_task_alloc(ptr @{{.+}}, i32 %{{.+}}, i32 1, i64 {{.+}}, i64 {{.+}}, ptr @_QPtest_team..omp_par)
More information about the flang-commits
mailing list