yutsumi created this revision.
yutsumi added reviewers: ABataev, jdoerfert, cchen.
Herald added subscribers: guansong, yaxunl.
Herald added a project: All.
yutsumi requested review of this revision.
Herald added subscribers: openmp-commits, cfe-commits, sstefan1.
Herald added projects: clang, OpenMP.
Make copyin clause on the following constructs work.
- parallel for
- parallel for simd
- parallel sections
Fixes https://github.com/llvm/llvm-project/issues/55547
Repository:
rG LLVM Github Monorepo
https://reviews.llvm.org/D132209
Files:
clang/lib/CodeGen/CGStmtOpenMP.cpp
openmp/runtime/test/parallel/omp_parallel_copyin_combined.c
Index: openmp/runtime/test/parallel/omp_parallel_copyin_combined.c
===================================================================
--- /dev/null
+++ openmp/runtime/test/parallel/omp_parallel_copyin_combined.c
@@ -0,0 +1,110 @@
+// RUN: %libomp-compile-and-run
+#include "omp_testsuite.h"
+
+#define N 100
+
+int x1, x2, x3, x4, x5;
+#pragma omp threadprivate(x1, x2, x3, x4, x5)
+
+int test_omp_parallel_copyin() {
+ int a[N];
+ x1 = 1;
+
+#pragma omp parallel copyin(x1)
+#pragma omp for
+ for (int i = 0; i < N; i++)
+ a[i] = i + x1;
+
+ int sum = 0;
+
+ for (int i = 0; i < N; i++)
+ sum += a[i];
+
+ return (sum == ((99 + 2 * x1) * 100) / 2);
+}
+
+int test_omp_parallel_for_copyin() {
+ int a[N];
+ x2 = 2;
+
+#pragma omp parallel for copyin(x2)
+ for (int i = 0; i < N; i++)
+ a[i] = i + x2;
+
+ int sum = 0;
+
+ for (int i = 0; i < N; i++)
+ sum += a[i];
+
+ return (sum == ((99 + 2 * x2) * 100) / 2);
+}
+
+int test_omp_parallel_for_simd_copyin() {
+ int a[N];
+ x3 = 3;
+
+#pragma omp parallel for simd copyin(x3)
+ for (int i = 0; i < N; i++)
+ a[i] = i + x3;
+
+ int sum = 0;
+
+ for (int i = 0; i < N; i++)
+ sum += a[i];
+
+ return (sum == ((99 + 2 * x3) * 100) / 2);
+}
+
+int test_omp_parallel_sections_copyin() {
+ int a = 0;
+ int b = 0;
+ x4 = 4;
+
+#pragma omp parallel sections copyin(x4)
+ {
+#pragma omp section
+ { a = x4; }
+
+#pragma omp section
+ { b = x4; }
+ }
+
+ return (a + b == x4 * 2);
+}
+
+int test_omp_parallel_master_copyin() {
+ int a[N];
+ x5 = 5;
+
+#pragma omp parallel master copyin(x5)
+ for (int i = 0; i < N; i++)
+ a[i] = i + x5;
+
+ int sum = 0;
+
+ for (int i = 0; i < N; i++)
+ sum += a[i];
+
+ return (sum == ((99 + 2 * x5) * 100) / 2);
+}
+
+int main() {
+ int num_failed = 0;
+
+ if (!test_omp_parallel_copyin())
+ num_failed++;
+
+ if (!test_omp_parallel_for_copyin())
+ num_failed++;
+
+ if (!test_omp_parallel_for_simd_copyin())
+ num_failed++;
+
+ if (!test_omp_parallel_sections_copyin())
+ num_failed++;
+
+ if (!test_omp_parallel_master_copyin())
+ num_failed++;
+
+ return num_failed;
+}
Index: clang/lib/CodeGen/CGStmtOpenMP.cpp
===================================================================
--- clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -1593,6 +1593,19 @@
const OMPExecutableDirective &,
llvm::SmallVectorImpl<llvm::Value *> &) {}
+static void emitOMPCopyinClause(CodeGenFunction &CGF,
+ const OMPExecutableDirective &S) {
+ bool Copyins = CGF.EmitOMPCopyinClause(S);
+ if (Copyins) {
+ // Emit implicit barrier to synchronize threads and avoid data races on
+ // propagation master's thread values of threadprivate variables to local
+ // instances of that variables of all other implicit threads.
+ CGF.CGM.getOpenMPRuntime().emitBarrierCall(
+ CGF, S.getBeginLoc(), OMPD_unknown, /*EmitChecks=*/false,
+ /*ForceSimpleCall=*/true);
+ }
+}
+
Address CodeGenFunction::OMPBuilderCBHelpers::getAddressOfLocalVariable(
CodeGenFunction &CGF, const VarDecl *VD) {
CodeGenModule &CGM = CGF.CGM;
@@ -1774,16 +1787,8 @@
auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &Action) {
Action.Enter(CGF);
OMPPrivateScope PrivateScope(CGF);
- bool Copyins = CGF.EmitOMPCopyinClause(S);
+ emitOMPCopyinClause(CGF, S);
(void)CGF.EmitOMPFirstprivateClause(S, PrivateScope);
- if (Copyins) {
- // Emit implicit barrier to synchronize threads and avoid data races on
- // propagation master's thread values of threadprivate variables to local
- // instances of that variables of all other implicit threads.
- CGF.CGM.getOpenMPRuntime().emitBarrierCall(
- CGF, S.getBeginLoc(), OMPD_unknown, /*EmitChecks=*/false,
- /*ForceSimpleCall=*/true);
- }
CGF.EmitOMPPrivateClause(S, PrivateScope);
CGF.EmitOMPReductionClauseInit(S, PrivateScope);
(void)PrivateScope.Privatize();
@@ -4352,6 +4357,7 @@
// directives: 'parallel' with 'for' directive.
auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &Action) {
Action.Enter(CGF);
+ emitOMPCopyinClause(CGF, S);
(void)emitWorksharingDirective(CGF, S, S.hasCancel());
};
{
@@ -4385,6 +4391,7 @@
// directives: 'parallel' with 'for' directive.
auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &Action) {
Action.Enter(CGF);
+ emitOMPCopyinClause(CGF, S);
(void)emitWorksharingDirective(CGF, S, /*HasCancel=*/false);
};
{
@@ -4419,16 +4426,8 @@
auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &Action) {
Action.Enter(CGF);
OMPPrivateScope PrivateScope(CGF);
- bool Copyins = CGF.EmitOMPCopyinClause(S);
+ emitOMPCopyinClause(CGF, S);
(void)CGF.EmitOMPFirstprivateClause(S, PrivateScope);
- if (Copyins) {
- // Emit implicit barrier to synchronize threads and avoid data races on
- // propagation master's thread values of threadprivate variables to local
- // instances of that variables of all other implicit threads.
- CGF.CGM.getOpenMPRuntime().emitBarrierCall(
- CGF, S.getBeginLoc(), OMPD_unknown, /*EmitChecks=*/false,
- /*ForceSimpleCall=*/true);
- }
CGF.EmitOMPPrivateClause(S, PrivateScope);
CGF.EmitOMPReductionClauseInit(S, PrivateScope);
(void)PrivateScope.Privatize();
@@ -4453,6 +4452,7 @@
// directives: 'parallel' with 'sections' directive.
auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &Action) {
Action.Enter(CGF);
+ emitOMPCopyinClause(CGF, S);
CGF.EmitSections(S);
};
{
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
