[Flang][OpenMP] Add Lowering support for taskloop reductions #166582
Closed
+39
−21
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Support for lowering the Reduction clause support already exists, so we can extend the support for taskloop to include reduction. As support for Reduction in taskloop was only added in OpenMP 5.0, the use of the Clause has been restricted to that version of OpenMP or greater.
llvmbot
added
flang
Member
|
@llvm/pr-subscribers-flang-openmp Author: Jack Styles (Stylie777) ChangesSupport for lowering the Reduction clause support already exists, so we can extend the support for taskloop to include reduction. As support for Reduction in taskloop was only added in OpenMP 5.0, the use of the Clause has been restricted to that version of OpenMP or greater. Full diff: https://github.com/llvm/llvm-project/pull/166582.diff 4 Files Affected:
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index ad456d89bc432..51170a39d272b 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -1763,21 +1763,22 @@ static void genTaskgroupClauses(
cp.processTaskReduction(loc, clauseOps, taskReductionSyms);
}
-static void genTaskloopClauses(lower::AbstractConverter &converter,
- semantics::SemanticsContext &semaCtx,
- lower::StatementContext &stmtCtx,
- const List<Clause> &clauses, mlir::Location loc,
- mlir::omp::TaskloopOperands &clauseOps) {
+static void genTaskloopClauses(
+ lower::AbstractConverter &converter, semantics::SemanticsContext &semaCtx,
+ lower::StatementContext &stmtCtx, const List<Clause> &clauses,
+ mlir::Location loc, mlir::omp::TaskloopOperands &clauseOps,
+ llvm::SmallVectorImpl<const semantics::Symbol *> &taskReductionSyms) {
ClauseProcessor cp(converter, semaCtx, clauses);
cp.processGrainsize(stmtCtx, clauseOps);
cp.processNumTasks(stmtCtx, clauseOps);
+ cp.processReduction(loc, clauseOps, taskReductionSyms);
cp.processTODO<clause::Allocate, clause::Collapse, clause::Default,
clause::Final, clause::If, clause::InReduction,
clause::Lastprivate, clause::Mergeable, clause::Nogroup,
- clause::Priority, clause::Reduction, clause::Shared,
- clause::Untied>(loc, llvm::omp::Directive::OMPD_taskloop);
+ clause::Priority, clause::Shared, clause::Untied>(
+ loc, llvm::omp::Directive::OMPD_taskloop);
}
static void genTaskwaitClauses(lower::AbstractConverter &converter,
@@ -2979,8 +2980,9 @@ static mlir::omp::TaskloopOp genStandaloneTaskloop(
lower::pft::Evaluation &eval, mlir::Location loc,
const ConstructQueue &queue, ConstructQueue::const_iterator item) {
mlir::omp::TaskloopOperands taskloopClauseOps;
+ llvm::SmallVector<const semantics::Symbol *> taskReductionSyms;
genTaskloopClauses(converter, semaCtx, stmtCtx, item->clauses, loc,
- taskloopClauseOps);
+ taskloopClauseOps, taskReductionSyms);
DataSharingProcessor dsp(converter, semaCtx, item->clauses, eval,
/*shouldCollectPreDeterminedSymbols=*/true,
enableDelayedPrivatization, symTable);
@@ -2994,6 +2996,8 @@ static mlir::omp::TaskloopOp genStandaloneTaskloop(
EntryBlockArgs taskloopArgs;
taskloopArgs.priv.syms = dsp.getDelayedPrivSymbols();
taskloopArgs.priv.vars = taskloopClauseOps.privateVars;
+ taskloopArgs.reduction.syms = taskReductionSyms;
+ taskloopArgs.reduction.vars = taskloopClauseOps.reductionVars;
auto taskLoopOp = genWrapperOp<mlir::omp::TaskloopOp>(
converter, loc, taskloopClauseOps, taskloopArgs);
diff --git a/flang/test/Lower/OpenMP/Todo/taskloop-reduction.f90 b/flang/test/Lower/OpenMP/Todo/taskloop-reduction.f90
deleted file mode 100644
index 0c16bd227257f..0000000000000
--- a/flang/test/Lower/OpenMP/Todo/taskloop-reduction.f90
+++ /dev/null
@@ -1,13 +0,0 @@
-! RUN: %not_todo_cmd bbc -emit-fir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
-! RUN: %not_todo_cmd %flang_fc1 -emit-fir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
-
-! CHECK: not yet implemented: Unhandled clause REDUCTION in TASKLOOP construct
-subroutine omp_taskloop_reduction()
- integer x
- x = 0
- !$omp taskloop reduction(+:x)
- do i = 1, 100
- x = x + 1
- end do
- !$omp end taskloop
-end subroutine omp_taskloop_reduction
diff --git a/flang/test/Lower/OpenMP/taskloop-reduction.f90 b/flang/test/Lower/OpenMP/taskloop-reduction.f90
new file mode 100644
index 0000000000000..4185a927366e7
--- /dev/null
+++ b/flang/test/Lower/OpenMP/taskloop-reduction.f90
@@ -0,0 +1,30 @@
+! This test checks the lowering of the reduction clause in the taskloop construct
+! RUN: bbc -emit-hlfir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
+! RUN %flang_fc1 -emit-hlfir -fopenmp -fopenmp-version=45 -o - %s 2>&1 | FileCheck %s --check-prefix=CHECK-VERSION
+
+! CHECK-VERSION: error: REDUCTION clause is not allowed on directive TASKLOOP in OpenMP v4.5, try -fopenmp-version=50
+
+! CHECK-LABEL: omp.private
+! CHECK-SAME: {type = private} @[[I_PRIVATE:.*]] : i32
+
+! CHECK-LABEL: func.func @_QPtest_reduction()
+! CHECK: %[[ALLOCA_A:.*]] = fir.alloca !fir.array<10xi32> {bindc_name = "a", uniq_name = "_QFtest_reductionEa"}
+! CHECK: %[[DECLARE_A:.*]]:2 = hlfir.declare %[[ALLOCA_A]](%2) {uniq_name = "_QFtest_reductionEa"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>)
+! CHECK: %[[ALLOCA_I:.*]] = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFtest_reductionEi"}
+! CHECK: %[[DECLARE_I:.*]]:2 = hlfir.declare %[[ALLOCA_I]] {uniq_name = "_QFtest_reductionEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: %[[ALLOCA_SUM_I:.*]] = fir.alloca i32 {bindc_name = "sum_i", uniq_name = "_QFtest_reductionEsum_i"}
+! CHECK: %[[DECLARE_SUM_I:.*]]:2 = hlfir.declare %[[ALLOCA_SUM_I]] {uniq_name = "_QFtest_reductionEsum_i"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+
+subroutine test_reduction
+ integer :: i, a(10), sum_i
+
+ ! CHECK: omp.taskloop
+ ! CHECK-SAME: private(@[[I_PRIVATE]] %[[DECLARE_I]]#0 -> %arg0 : !fir.ref<i32>) reduction(@add_reduction_i32 %[[DECLARE_SUM_I]]#0 -> %arg1 : !fir.ref<i32>) {
+ !$omp taskloop reduction (+:sum_i)
+ do i = 1,10
+ sum_i = sum_i + i
+ end do
+ !$omp end taskloop
+
+end subroutine
\ No newline at end of file
diff --git a/llvm/include/llvm/Frontend/OpenMP/OMP.td b/llvm/include/llvm/Frontend/OpenMP/OMP.td
index 208609f64f418..0afae8a013bd6 100644
--- a/llvm/include/llvm/Frontend/OpenMP/OMP.td
+++ b/llvm/include/llvm/Frontend/OpenMP/OMP.td
@@ -1284,17 +1284,16 @@ def OMP_TaskGroup : Directive<[Spelling<"taskgroup">]> {
let category = CA_Executable;
}
def OMP_TaskLoop : Directive<[Spelling<"taskloop">]> {
- let allowedClauses = [
- VersionedClause<OMPC_Allocate>,
- VersionedClause<OMPC_FirstPrivate>,
- VersionedClause<OMPC_InReduction>,
- VersionedClause<OMPC_LastPrivate>,
- VersionedClause<OMPC_Mergeable>,
- VersionedClause<OMPC_NoGroup>,
- VersionedClause<OMPC_Private>,
- VersionedClause<OMPC_Reduction>,
- VersionedClause<OMPC_Shared>,
- VersionedClause<OMPC_Untied>,
+ let allowedClauses = [VersionedClause<OMPC_Allocate>,
+ VersionedClause<OMPC_FirstPrivate>,
+ VersionedClause<OMPC_InReduction>,
+ VersionedClause<OMPC_LastPrivate>,
+ VersionedClause<OMPC_Mergeable>,
+ VersionedClause<OMPC_NoGroup>,
+ VersionedClause<OMPC_Private>,
+ VersionedClause<OMPC_Reduction, 50>,
+ VersionedClause<OMPC_Shared>,
+ VersionedClause<OMPC_Untied>,
];
let allowedOnceClauses = [
VersionedClause<OMPC_Collapse>,
|
Member
|
@llvm/pr-subscribers-flang-fir-hlfir Author: Jack Styles (Stylie777) ChangesSupport for lowering the Reduction clause support already exists, so we can extend the support for taskloop to include reduction. As support for Reduction in taskloop was only added in OpenMP 5.0, the use of the Clause has been restricted to that version of OpenMP or greater. Full diff: https://github.com/llvm/llvm-project/pull/166582.diff 4 Files Affected:
diff --git a/flang/lib/Lower/OpenMP/OpenMP.cpp b/flang/lib/Lower/OpenMP/OpenMP.cpp
index ad456d89bc432..51170a39d272b 100644
--- a/flang/lib/Lower/OpenMP/OpenMP.cpp
+++ b/flang/lib/Lower/OpenMP/OpenMP.cpp
@@ -1763,21 +1763,22 @@ static void genTaskgroupClauses(
cp.processTaskReduction(loc, clauseOps, taskReductionSyms);
}
-static void genTaskloopClauses(lower::AbstractConverter &converter,
- semantics::SemanticsContext &semaCtx,
- lower::StatementContext &stmtCtx,
- const List<Clause> &clauses, mlir::Location loc,
- mlir::omp::TaskloopOperands &clauseOps) {
+static void genTaskloopClauses(
+ lower::AbstractConverter &converter, semantics::SemanticsContext &semaCtx,
+ lower::StatementContext &stmtCtx, const List<Clause> &clauses,
+ mlir::Location loc, mlir::omp::TaskloopOperands &clauseOps,
+ llvm::SmallVectorImpl<const semantics::Symbol *> &taskReductionSyms) {
ClauseProcessor cp(converter, semaCtx, clauses);
cp.processGrainsize(stmtCtx, clauseOps);
cp.processNumTasks(stmtCtx, clauseOps);
+ cp.processReduction(loc, clauseOps, taskReductionSyms);
cp.processTODO<clause::Allocate, clause::Collapse, clause::Default,
clause::Final, clause::If, clause::InReduction,
clause::Lastprivate, clause::Mergeable, clause::Nogroup,
- clause::Priority, clause::Reduction, clause::Shared,
- clause::Untied>(loc, llvm::omp::Directive::OMPD_taskloop);
+ clause::Priority, clause::Shared, clause::Untied>(
+ loc, llvm::omp::Directive::OMPD_taskloop);
}
static void genTaskwaitClauses(lower::AbstractConverter &converter,
@@ -2979,8 +2980,9 @@ static mlir::omp::TaskloopOp genStandaloneTaskloop(
lower::pft::Evaluation &eval, mlir::Location loc,
const ConstructQueue &queue, ConstructQueue::const_iterator item) {
mlir::omp::TaskloopOperands taskloopClauseOps;
+ llvm::SmallVector<const semantics::Symbol *> taskReductionSyms;
genTaskloopClauses(converter, semaCtx, stmtCtx, item->clauses, loc,
- taskloopClauseOps);
+ taskloopClauseOps, taskReductionSyms);
DataSharingProcessor dsp(converter, semaCtx, item->clauses, eval,
/*shouldCollectPreDeterminedSymbols=*/true,
enableDelayedPrivatization, symTable);
@@ -2994,6 +2996,8 @@ static mlir::omp::TaskloopOp genStandaloneTaskloop(
EntryBlockArgs taskloopArgs;
taskloopArgs.priv.syms = dsp.getDelayedPrivSymbols();
taskloopArgs.priv.vars = taskloopClauseOps.privateVars;
+ taskloopArgs.reduction.syms = taskReductionSyms;
+ taskloopArgs.reduction.vars = taskloopClauseOps.reductionVars;
auto taskLoopOp = genWrapperOp<mlir::omp::TaskloopOp>(
converter, loc, taskloopClauseOps, taskloopArgs);
diff --git a/flang/test/Lower/OpenMP/Todo/taskloop-reduction.f90 b/flang/test/Lower/OpenMP/Todo/taskloop-reduction.f90
deleted file mode 100644
index 0c16bd227257f..0000000000000
--- a/flang/test/Lower/OpenMP/Todo/taskloop-reduction.f90
+++ /dev/null
@@ -1,13 +0,0 @@
-! RUN: %not_todo_cmd bbc -emit-fir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
-! RUN: %not_todo_cmd %flang_fc1 -emit-fir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
-
-! CHECK: not yet implemented: Unhandled clause REDUCTION in TASKLOOP construct
-subroutine omp_taskloop_reduction()
- integer x
- x = 0
- !$omp taskloop reduction(+:x)
- do i = 1, 100
- x = x + 1
- end do
- !$omp end taskloop
-end subroutine omp_taskloop_reduction
diff --git a/flang/test/Lower/OpenMP/taskloop-reduction.f90 b/flang/test/Lower/OpenMP/taskloop-reduction.f90
new file mode 100644
index 0000000000000..4185a927366e7
--- /dev/null
+++ b/flang/test/Lower/OpenMP/taskloop-reduction.f90
@@ -0,0 +1,30 @@
+! This test checks the lowering of the reduction clause in the taskloop construct
+! RUN: bbc -emit-hlfir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
+! RUN: %flang_fc1 -emit-hlfir -fopenmp -fopenmp-version=50 -o - %s 2>&1 | FileCheck %s
+! RUN %flang_fc1 -emit-hlfir -fopenmp -fopenmp-version=45 -o - %s 2>&1 | FileCheck %s --check-prefix=CHECK-VERSION
+
+! CHECK-VERSION: error: REDUCTION clause is not allowed on directive TASKLOOP in OpenMP v4.5, try -fopenmp-version=50
+
+! CHECK-LABEL: omp.private
+! CHECK-SAME: {type = private} @[[I_PRIVATE:.*]] : i32
+
+! CHECK-LABEL: func.func @_QPtest_reduction()
+! CHECK: %[[ALLOCA_A:.*]] = fir.alloca !fir.array<10xi32> {bindc_name = "a", uniq_name = "_QFtest_reductionEa"}
+! CHECK: %[[DECLARE_A:.*]]:2 = hlfir.declare %[[ALLOCA_A]](%2) {uniq_name = "_QFtest_reductionEa"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>)
+! CHECK: %[[ALLOCA_I:.*]] = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFtest_reductionEi"}
+! CHECK: %[[DECLARE_I:.*]]:2 = hlfir.declare %[[ALLOCA_I]] {uniq_name = "_QFtest_reductionEi"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+! CHECK: %[[ALLOCA_SUM_I:.*]] = fir.alloca i32 {bindc_name = "sum_i", uniq_name = "_QFtest_reductionEsum_i"}
+! CHECK: %[[DECLARE_SUM_I:.*]]:2 = hlfir.declare %[[ALLOCA_SUM_I]] {uniq_name = "_QFtest_reductionEsum_i"} : (!fir.ref<i32>) -> (!fir.ref<i32>, !fir.ref<i32>)
+
+subroutine test_reduction
+ integer :: i, a(10), sum_i
+
+ ! CHECK: omp.taskloop
+ ! CHECK-SAME: private(@[[I_PRIVATE]] %[[DECLARE_I]]#0 -> %arg0 : !fir.ref<i32>) reduction(@add_reduction_i32 %[[DECLARE_SUM_I]]#0 -> %arg1 : !fir.ref<i32>) {
+ !$omp taskloop reduction (+:sum_i)
+ do i = 1,10
+ sum_i = sum_i + i
+ end do
+ !$omp end taskloop
+
+end subroutine
\ No newline at end of file
diff --git a/llvm/include/llvm/Frontend/OpenMP/OMP.td b/llvm/include/llvm/Frontend/OpenMP/OMP.td
index 208609f64f418..0afae8a013bd6 100644
--- a/llvm/include/llvm/Frontend/OpenMP/OMP.td
+++ b/llvm/include/llvm/Frontend/OpenMP/OMP.td
@@ -1284,17 +1284,16 @@ def OMP_TaskGroup : Directive<[Spelling<"taskgroup">]> {
let category = CA_Executable;
}
def OMP_TaskLoop : Directive<[Spelling<"taskloop">]> {
- let allowedClauses = [
- VersionedClause<OMPC_Allocate>,
- VersionedClause<OMPC_FirstPrivate>,
- VersionedClause<OMPC_InReduction>,
- VersionedClause<OMPC_LastPrivate>,
- VersionedClause<OMPC_Mergeable>,
- VersionedClause<OMPC_NoGroup>,
- VersionedClause<OMPC_Private>,
- VersionedClause<OMPC_Reduction>,
- VersionedClause<OMPC_Shared>,
- VersionedClause<OMPC_Untied>,
+ let allowedClauses = [VersionedClause<OMPC_Allocate>,
+ VersionedClause<OMPC_FirstPrivate>,
+ VersionedClause<OMPC_InReduction>,
+ VersionedClause<OMPC_LastPrivate>,
+ VersionedClause<OMPC_Mergeable>,
+ VersionedClause<OMPC_NoGroup>,
+ VersionedClause<OMPC_Private>,
+ VersionedClause<OMPC_Reduction, 50>,
+ VersionedClause<OMPC_Shared>,
+ VersionedClause<OMPC_Untied>,
];
let allowedOnceClauses = [
VersionedClause<OMPC_Collapse>,
|
…_taskloop_reductions
While this was introduced in OpenMP 5.0 for taskloop, it has not previously been restricted and is irrelevant to adding support in flang. If we want to restrict this, we should do it in a separate patch.
Contributor
Author
|
Abandoning in favour of #166751 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Support for lowering the Reduction clause support already exists, so we can extend the support for taskloop to include reduction.
As support for Reduction in taskloop was only added in OpenMP 5.0, the use of the Clause has been restricted to that version of OpenMP or greater.