jyu2 updated this revision to Diff 536394.
jyu2 marked an inline comment as done.
jyu2 added a comment.
Thanks Alexey's review. This address Alexey's comments.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D154180/new/
https://reviews.llvm.org/D154180
Files:
clang/lib/CodeGen/CGOpenMPRuntime.cpp
clang/lib/CodeGen/CGOpenMPRuntime.h
clang/lib/CodeGen/CGStmtOpenMP.cpp
clang/test/OpenMP/ordered_doacross_codegen.c
Index: clang/test/OpenMP/ordered_doacross_codegen.c
===================================================================
--- clang/test/OpenMP/ordered_doacross_codegen.c
+++ clang/test/OpenMP/ordered_doacross_codegen.c
@@ -2,13 +2,21 @@
// RUN: %clang_cc1 -fopenmp -triple x86_64-unknown-unknown -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp -triple x86_64-unknown-unknown -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefixes=CHECK,CHECK-NORMAL
+// RUN: %clang_cc1 -verify -fopenmp -triple x86_64-unknown-unknown -emit-llvm %s -o - -fopenmp-version=52 | FileCheck %s --check-prefixes=CHECK,CHECK-NORMAL
// RUN: %clang_cc1 -verify -fopenmp -fopenmp-enable-irbuilder -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-IRBUILDER
+// RUN: %clang_cc1 -verify -fopenmp -fopenmp-enable-irbuilder -triple x86_64-unknown-unknown -fopenmp-version=52 -emit-llvm %s -o - | FileCheck %s --check-prefixes=CHECK,CHECK-IRBUILDER
// RUN: %clang_cc1 -fopenmp -fopenmp-enable-irbuilder -triple x86_64-unknown-unknown -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp -fopenmp-enable-irbuilder -triple x86_64-unknown-unknown -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefixes=CHECK,CHECK-IRBUILDER
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=52 -fopenmp-enable-irbuilder -triple x86_64-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp -fopenmp-version=52 -fopenmp-enable-irbuilder -triple x86_64-unknown-unknown -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefixes=CHECK,CHECK-IRBUILDER
+
// RUN: %clang_cc1 -verify -fopenmp-simd -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -verify -fopenmp-simd -fopenmp-version=52 -triple x86_64-unknown-unknown -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -fopenmp-simd -triple x86_64-unknown-unknown -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -triple x86_64-unknown-unknown -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=52 -triple x86_64-unknown-unknown -emit-pch -o %t %s
+// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=52 -triple x86_64-unknown-unknown -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
// expected-no-diagnostics
@@ -51,7 +59,11 @@
// CHECK-NORMAL-NEXT: call void @__kmpc_doacross_post(ptr [[IDENT]], i32 [[GTID]], ptr [[TMP]])
// CHECK-IRBUILDER-NEXT: [[GTID1:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[IDENT:@.+]])
// CHECK-IRBUILDER-NEXT: call void @__kmpc_doacross_post(ptr [[IDENT]], i32 [[GTID1]], ptr [[TMP]])
+#if _OPENMP >= 202111
+#pragma omp ordered doacross(source:)
+#else
#pragma omp ordered depend(source)
+#endif
c[i] = c[i] + 1;
foo();
// CHECK: call void @foo()
@@ -66,7 +78,11 @@
// CHECK-NORMAL-NEXT: call void @__kmpc_doacross_wait(ptr [[IDENT]], i32 [[GTID]], ptr [[TMP]])
// CHECK-IRBUILDER-NEXT: [[GTID2:%.+]] = call i32 @__kmpc_global_thread_num(ptr [[IDENT:@.+]])
// CHECK-IRBUILDER-NEXT: call void @__kmpc_doacross_wait(ptr [[IDENT]], i32 [[GTID2]], ptr [[TMP]])
+#if _OPENMP >= 202111
+#pragma omp ordered doacross(sink : i - 2)
+#else
#pragma omp ordered depend(sink : i - 2)
+#endif
d[i] = a[i - 2];
}
// CHECK: call void @__kmpc_for_static_fini(
Index: clang/lib/CodeGen/CGStmtOpenMP.cpp
===================================================================
--- clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -5839,37 +5839,50 @@
return Fn;
}
+template <typename T>
+static void emitRestoreIP(CodeGenFunction &CGF, const T *C,
+ llvm::OpenMPIRBuilder::InsertPointTy AllocaIP,
+ llvm::OpenMPIRBuilder &OMPBuilder) {
+
+ unsigned NumLoops = C->getNumLoops();
+ QualType Int64Ty = CGF.CGM.getContext().getIntTypeForBitwidth(
+ /*DestWidth=*/64, /*Signed=*/1);
+ llvm::SmallVector<llvm::Value *> StoreValues;
+ for (unsigned I = 0; I < NumLoops; I++) {
+ const Expr *CounterVal = C->getLoopData(I);
+ assert(CounterVal);
+ llvm::Value *StoreValue = CGF.EmitScalarConversion(
+ CGF.EmitScalarExpr(CounterVal), CounterVal->getType(), Int64Ty,
+ CounterVal->getExprLoc());
+ StoreValues.emplace_back(StoreValue);
+ }
+ const auto DOC = dyn_cast<OMPDoacrossClause>(C);
+ const auto DC = dyn_cast<OMPDependClause>(C);
+ bool IsDependSource = false;
+ if ((DC && DC->getDependencyKind() == OMPC_DEPEND_source) ||
+ (DOC && DOC->getDependenceType() == OMPC_DOACROSS_source))
+ IsDependSource = true;
+ CGF.Builder.restoreIP(
+ OMPBuilder.createOrderedDepend(CGF.Builder, AllocaIP, NumLoops,
+ StoreValues, ".cnt.addr", IsDependSource));
+}
+
void CodeGenFunction::EmitOMPOrderedDirective(const OMPOrderedDirective &S) {
if (CGM.getLangOpts().OpenMPIRBuilder) {
llvm::OpenMPIRBuilder &OMPBuilder = CGM.getOpenMPRuntime().getOMPBuilder();
using InsertPointTy = llvm::OpenMPIRBuilder::InsertPointTy;
- if (S.hasClausesOfKind<OMPDependClause>()) {
+ if (S.hasClausesOfKind<OMPDependClause>() ||
+ S.hasClausesOfKind<OMPDoacrossClause>()) {
// The ordered directive with depend clause.
- assert(!S.hasAssociatedStmt() &&
- "No associated statement must be in ordered depend construct.");
+ assert(!S.hasAssociatedStmt() && "No associated statement must be in "
+ "ordered depend|doacross construct.");
InsertPointTy AllocaIP(AllocaInsertPt->getParent(),
AllocaInsertPt->getIterator());
- for (const auto *DC : S.getClausesOfKind<OMPDependClause>()) {
- unsigned NumLoops = DC->getNumLoops();
- QualType Int64Ty = CGM.getContext().getIntTypeForBitwidth(
- /*DestWidth=*/64, /*Signed=*/1);
- llvm::SmallVector<llvm::Value *> StoreValues;
- for (unsigned I = 0; I < NumLoops; I++) {
- const Expr *CounterVal = DC->getLoopData(I);
- assert(CounterVal);
- llvm::Value *StoreValue = EmitScalarConversion(
- EmitScalarExpr(CounterVal), CounterVal->getType(), Int64Ty,
- CounterVal->getExprLoc());
- StoreValues.emplace_back(StoreValue);
- }
- bool IsDependSource = false;
- if (DC->getDependencyKind() == OMPC_DEPEND_source)
- IsDependSource = true;
- Builder.restoreIP(OMPBuilder.createOrderedDepend(
- Builder, AllocaIP, NumLoops, StoreValues, ".cnt.addr",
- IsDependSource));
- }
+ for (const auto *DC : S.getClausesOfKind<OMPDependClause>())
+ emitRestoreIP(*this, DC, AllocaIP, OMPBuilder);
+ for (const auto *DC : S.getClausesOfKind<OMPDoacrossClause>())
+ emitRestoreIP(*this, DC, AllocaIP, OMPBuilder);
} else {
// The ordered directive with threads or simd clause, or without clause.
// Without clause, it behaves as if the threads clause is specified.
@@ -5916,6 +5929,13 @@
CGM.getOpenMPRuntime().emitDoacrossOrdered(*this, DC);
return;
}
+ if (S.hasClausesOfKind<OMPDoacrossClause>()) {
+ assert(!S.hasAssociatedStmt() &&
+ "No associated statement must be in ordered doacross construct.");
+ for (const auto *DC : S.getClausesOfKind<OMPDoacrossClause>())
+ CGM.getOpenMPRuntime().emitDoacrossOrdered(*this, DC);
+ return;
+ }
const auto *C = S.getSingleClause<OMPSIMDClause>();
auto &&CodeGen = [&S, C, this](CodeGenFunction &CGF,
PrePostActionTy &Action) {
Index: clang/lib/CodeGen/CGOpenMPRuntime.h
===================================================================
--- clang/lib/CodeGen/CGOpenMPRuntime.h
+++ clang/lib/CodeGen/CGOpenMPRuntime.h
@@ -327,13 +327,6 @@
bool IsOffloadEntry,
const RegionCodeGenTy &CodeGen);
- /// Emits object of ident_t type with info for source location.
- /// \param Flags Flags for OpenMP location.
- /// \param EmitLoc emit source location with debug-info is off.
- ///
- llvm::Value *emitUpdateLocation(CodeGenFunction &CGF, SourceLocation Loc,
- unsigned Flags = 0, bool EmitLoc = false);
-
/// Emit the number of teams for a target directive. Inspect the num_teams
/// clause associated with a teams construct combined or closely nested
/// with the target directive.
@@ -366,10 +359,6 @@
/// Returns pointer to ident_t type.
llvm::Type *getIdentTyPointerTy();
- /// Gets thread id value for the current thread.
- ///
- llvm::Value *getThreadID(CodeGenFunction &CGF, SourceLocation Loc);
-
/// Get the function name of an outlined region.
std::string getOutlinedHelperName(StringRef Name) const;
std::string getOutlinedHelperName(CodeGenFunction &CGF) const;
@@ -1487,6 +1476,11 @@
virtual void emitDoacrossOrdered(CodeGenFunction &CGF,
const OMPDependClause *C);
+ /// Emit code for doacross ordered directive with 'doacross' clause.
+ /// \param C 'doacross' clause with 'sink|source' dependence type.
+ virtual void emitDoacrossOrdered(CodeGenFunction &CGF,
+ const OMPDoacrossClause *C);
+
/// Translates the native parameter of outlined function if this is required
/// for target.
/// \param FD Field decl from captured record for the parameter.
@@ -1653,6 +1647,17 @@
/// Returns true if the variable is a local variable in untied task.
bool isLocalVarInUntiedTask(CodeGenFunction &CGF, const VarDecl *VD) const;
+
+ /// Gets thread id value for the current thread.
+ ///
+ llvm::Value *getThreadID(CodeGenFunction &CGF, SourceLocation Loc);
+
+ /// Emits object of ident_t type with info for source location.
+ /// \param Flags Flags for OpenMP location.
+ /// \param EmitLoc emit source location with debug-info is off.
+ ///
+ llvm::Value *emitUpdateLocation(CodeGenFunction &CGF, SourceLocation Loc,
+ unsigned Flags = 0, bool EmitLoc = false);
};
/// Class supports emissionof SIMD-only code.
@@ -2240,6 +2245,11 @@
void emitDoacrossOrdered(CodeGenFunction &CGF,
const OMPDependClause *C) override;
+ /// Emit code for doacross ordered directive with 'doacross' clause.
+ /// \param C 'doacross' clause with 'sink|source' dependence type.
+ void emitDoacrossOrdered(CodeGenFunction &CGF,
+ const OMPDoacrossClause *C) override;
+
/// Translates the native parameter of outlined function if this is required
/// for target.
/// \param FD Field decl from captured record for the parameter.
Index: clang/lib/CodeGen/CGOpenMPRuntime.cpp
===================================================================
--- clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -11379,8 +11379,9 @@
llvm::ArrayRef(FiniArgs));
}
-void CGOpenMPRuntime::emitDoacrossOrdered(CodeGenFunction &CGF,
- const OMPDependClause *C) {
+template <typename T>
+static void EmitDoacrossOrdered(CGOpenMPRuntime &RT, CodeGenFunction &CGF,
+ CodeGenModule &CGM, const T *C) {
QualType Int64Ty =
CGM.getContext().getIntTypeForBitwidth(/*DestWidth=*/64, /*Signed=*/1);
llvm::APInt Size(/*numBits=*/32, C->getNumLoops());
@@ -11397,21 +11398,36 @@
/*Volatile=*/false, Int64Ty);
}
llvm::Value *Args[] = {
- emitUpdateLocation(CGF, C->getBeginLoc()),
- getThreadID(CGF, C->getBeginLoc()),
+ RT.emitUpdateLocation(CGF, C->getBeginLoc()),
+ RT.getThreadID(CGF, C->getBeginLoc()),
CGF.Builder.CreateConstArrayGEP(CntAddr, 0).getPointer()};
llvm::FunctionCallee RTLFn;
- if (C->getDependencyKind() == OMPC_DEPEND_source) {
+ llvm::OpenMPIRBuilder &OMPBuilder = CGM.getOpenMPRuntime().getOMPBuilder();
+ const OMPDependClause *DepC = dyn_cast<OMPDependClause>(C);
+ const OMPDoacrossClause *DoC = dyn_cast<OMPDoacrossClause>(C);
+ if ((DoC && DoC->getDependenceType() == OMPC_DOACROSS_source) ||
+ (DepC && DepC->getDependencyKind() == OMPC_DEPEND_source)) {
RTLFn = OMPBuilder.getOrCreateRuntimeFunction(CGM.getModule(),
OMPRTL___kmpc_doacross_post);
} else {
- assert(C->getDependencyKind() == OMPC_DEPEND_sink);
+ assert((DoC && DoC->getDependenceType() == OMPC_DOACROSS_sink) ||
+ (DepC && DepC->getDependencyKind() == OMPC_DEPEND_sink));
RTLFn = OMPBuilder.getOrCreateRuntimeFunction(CGM.getModule(),
OMPRTL___kmpc_doacross_wait);
}
CGF.EmitRuntimeCall(RTLFn, Args);
}
+void CGOpenMPRuntime::emitDoacrossOrdered(CodeGenFunction &CGF,
+ const OMPDependClause *CL) {
+ return EmitDoacrossOrdered<OMPDependClause>(*this, CGF, CGM, CL);
+}
+
+void CGOpenMPRuntime::emitDoacrossOrdered(CodeGenFunction &CGF,
+ const OMPDoacrossClause *CL) {
+ return EmitDoacrossOrdered<OMPDoacrossClause>(*this, CGF, CGM, CL);
+}
+
void CGOpenMPRuntime::emitCall(CodeGenFunction &CGF, SourceLocation Loc,
llvm::FunctionCallee Callee,
ArrayRef<llvm::Value *> Args) const {
@@ -12398,6 +12414,11 @@
llvm_unreachable("Not supported in SIMD-only mode");
}
+void CGOpenMPSIMDRuntime::emitDoacrossOrdered(CodeGenFunction &CGF,
+ const OMPDoacrossClause *C) {
+ llvm_unreachable("Not supported in SIMD-only mode");
+}
+
const VarDecl *
CGOpenMPSIMDRuntime::translateParameter(const FieldDecl *FD,
const VarDecl *NativeParam) const {
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
