[OPENMP50]Track changes of lastprivate conditional in parallel-based
regions with reductions, lastprivates or linears clauses.
If the lastprivate conditional variable is updated in inner parallel
region with reduction, lastprivate or linear clause, the value must be
considred as a candidate for lastprivate conditional. Also, tracking in
inner parallel regions is not required.
diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
index 896cf37..aecf150 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -11357,25 +11357,7 @@
}
}
Data.IVLVal = IVLVal;
- // In simd only mode or for simd directives no need to generate threadprivate
- // references for the loop iteration counter, we can use the original one
- // since outlining cannot happen in simd regions.
- if (CGF.getLangOpts().OpenMPSimd ||
- isOpenMPSimdDirective(S.getDirectiveKind())) {
- Data.UseOriginalIV = true;
- return;
- }
- PresumedLoc PLoc =
- CGM.getContext().getSourceManager().getPresumedLoc(S.getBeginLoc());
- assert(PLoc.isValid() && "Source location is expected to be always valid.");
-
- llvm::sys::fs::UniqueID ID;
- if (auto EC = llvm::sys::fs::getUniqueID(PLoc.getFilename(), ID))
- CGM.getDiags().Report(diag::err_cannot_open_file)
- << PLoc.getFilename() << EC.message();
- Data.IVName = CGM.getOpenMPRuntime().getName(
- {"pl_cond", llvm::utostr(ID.getDevice()), llvm::utostr(ID.getFile()),
- llvm::utostr(PLoc.getLine()), llvm::utostr(PLoc.getColumn()), "iv"});
+ Data.CGF = &CGF;
}
CGOpenMPRuntime::LastprivateConditionalRAII::~LastprivateConditionalRAII() {
@@ -11384,27 +11366,6 @@
CGM.getOpenMPRuntime().LastprivateConditionalStack.pop_back();
}
-void CGOpenMPRuntime::initLastprivateConditionalCounter(
- CodeGenFunction &CGF, const OMPExecutableDirective &S) {
- if (CGM.getLangOpts().OpenMPSimd ||
- !llvm::any_of(S.getClausesOfKind<OMPLastprivateClause>(),
- [](const OMPLastprivateClause *C) {
- return C->getKind() == OMPC_LASTPRIVATE_conditional;
- }))
- return;
- const CGOpenMPRuntime::LastprivateConditionalData &Data =
- LastprivateConditionalStack.back();
- if (Data.UseOriginalIV)
- return;
- // Global loop counter. Required to handle inner parallel-for regions.
- // global_iv = iv;
- Address GlobIVAddr = CGM.getOpenMPRuntime().getAddrOfArtificialThreadPrivate(
- CGF, Data.IVLVal.getType(), Data.IVName);
- LValue GlobIVLVal = CGF.MakeAddrLValue(GlobIVAddr, Data.IVLVal.getType());
- llvm::Value *IVVal = CGF.EmitLoadOfScalar(Data.IVLVal, S.getBeginLoc());
- CGF.EmitStoreOfScalar(IVVal, GlobIVLVal);
-}
-
namespace {
/// Checks if the lastprivate conditional variable is referenced in LHS.
class LastprivateConditionalRefChecker final
@@ -11415,9 +11376,7 @@
const Decl *FoundD = nullptr;
StringRef UniqueDeclName;
LValue IVLVal;
- StringRef IVName;
SourceLocation Loc;
- bool UseOriginalIV = false;
public:
bool VisitDeclRefExpr(const DeclRefExpr *E) {
@@ -11430,8 +11389,6 @@
FoundD = E->getDecl()->getCanonicalDecl();
UniqueDeclName = It->getSecond();
IVLVal = D.IVLVal;
- IVName = D.IVName;
- UseOriginalIV = D.UseOriginalIV;
break;
}
return FoundE == E;
@@ -11448,8 +11405,6 @@
FoundD = E->getMemberDecl()->getCanonicalDecl();
UniqueDeclName = It->getSecond();
IVLVal = D.IVLVal;
- IVName = D.IVName;
- UseOriginalIV = D.UseOriginalIV;
break;
}
return FoundE == E;
@@ -11470,17 +11425,17 @@
CodeGenFunction &CGF,
ArrayRef<CGOpenMPRuntime::LastprivateConditionalData> LPM)
: CGF(CGF), LPM(LPM) {}
- std::tuple<const Expr *, const Decl *, StringRef, LValue, StringRef, bool>
+ std::tuple<const Expr *, const Decl *, StringRef, LValue>
getFoundData() const {
- return std::make_tuple(FoundE, FoundD, UniqueDeclName, IVLVal, IVName,
- UseOriginalIV);
+ return std::make_tuple(FoundE, FoundD, UniqueDeclName, IVLVal);
}
};
} // namespace
void CGOpenMPRuntime::checkAndEmitLastprivateConditional(CodeGenFunction &CGF,
const Expr *LHS) {
- if (CGF.getLangOpts().OpenMP < 50)
+ if (CGF.getLangOpts().OpenMP < 50 || LastprivateConditionalStack.empty() ||
+ LastprivateConditionalStack.back().CGF != &CGF)
return;
LastprivateConditionalRefChecker Checker(CGF, LastprivateConditionalStack);
if (!Checker.Visit(LHS))
@@ -11489,10 +11444,7 @@
const Decl *FoundD;
StringRef UniqueDeclName;
LValue IVLVal;
- StringRef IVName;
- bool UseOriginalIV;
- std::tie(FoundE, FoundD, UniqueDeclName, IVLVal, IVName, UseOriginalIV) =
- Checker.getFoundData();
+ std::tie(FoundE, FoundD, UniqueDeclName, IVLVal) = Checker.getFoundData();
// Last updated loop counter for the lastprivate conditional var.
// int<xx> last_iv = 0;
@@ -11517,11 +11469,6 @@
// Global loop counter. Required to handle inner parallel-for regions.
// global_iv
- if (!UseOriginalIV) {
- Address IVAddr =
- getAddrOfArtificialThreadPrivate(CGF, IVLVal.getType(), IVName);
- IVLVal = CGF.MakeAddrLValue(IVAddr, IVLVal.getType());
- }
llvm::Value *IVVal = CGF.EmitLoadOfScalar(IVLVal, FoundE->getExprLoc());
// #pragma omp critical(a)
diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.h b/clang/lib/CodeGen/CGOpenMPRuntime.h
index 8159f5e..dbbf253 100644
--- a/clang/lib/CodeGen/CGOpenMPRuntime.h
+++ b/clang/lib/CodeGen/CGOpenMPRuntime.h
@@ -233,11 +233,7 @@
llvm::SmallDenseMap<CanonicalDeclPtr<const Decl>, SmallString<16>>
DeclToUniqeName;
LValue IVLVal;
- SmallString<16> IVName;
- /// True if original lvalue for loop counter can be used in codegen (simd
- /// region or simd only mode) and no need to create threadprivate
- /// references.
- bool UseOriginalIV = false;
+ CodeGenFunction *CGF = nullptr;
};
/// Manages list of lastprivate conditional decls for the specified directive.
class LastprivateConditionalRAII {
@@ -1692,11 +1688,6 @@
/// current context.
bool isNontemporalDecl(const ValueDecl *VD) const;
- /// Initializes global counter for lastprivate conditional.
- virtual void
- initLastprivateConditionalCounter(CodeGenFunction &CGF,
- const OMPExecutableDirective &S);
-
/// Checks if the provided \p LVal is lastprivate conditional and emits the
/// code to update the value of the original variable.
/// \code
diff --git a/clang/lib/CodeGen/CGStmtOpenMP.cpp b/clang/lib/CodeGen/CGStmtOpenMP.cpp
index f44405a..0e41d52 100644
--- a/clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ b/clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -18,6 +18,7 @@
#include "clang/AST/ASTContext.h"
#include "clang/AST/Attr.h"
#include "clang/AST/DeclOpenMP.h"
+#include "clang/AST/OpenMPClause.h"
#include "clang/AST/Stmt.h"
#include "clang/AST/StmtOpenMP.h"
#include "clang/Basic/PrettyStackTrace.h"
@@ -1332,6 +1333,19 @@
CGF.GenerateOpenMPCapturedVars(*CS, CapturedVars);
CGF.CGM.getOpenMPRuntime().emitParallelCall(CGF, S.getBeginLoc(), OutlinedFn,
CapturedVars, IfCond);
+ // Check for outer lastprivate conditional update.
+ for (const auto *C : S.getClausesOfKind<OMPReductionClause>()) {
+ for (const Expr *Ref : C->varlists())
+ CGF.CGM.getOpenMPRuntime().checkAndEmitLastprivateConditional(CGF, Ref);
+ }
+ for (const auto *C : S.getClausesOfKind<OMPLastprivateClause>()) {
+ for (const Expr *Ref : C->varlists())
+ CGF.CGM.getOpenMPRuntime().checkAndEmitLastprivateConditional(CGF, Ref);
+ }
+ for (const auto *C : S.getClausesOfKind<OMPLinearClause>()) {
+ for (const Expr *Ref : C->varlists())
+ CGF.CGM.getOpenMPRuntime().checkAndEmitLastprivateConditional(CGF, Ref);
+ }
}
static void emitEmptyBoundParameters(CodeGenFunction &,
@@ -1890,7 +1904,6 @@
static void emitOMPLoopBodyWithStopPoint(CodeGenFunction &CGF,
const OMPLoopDirective &S,
CodeGenFunction::JumpDest LoopExit) {
- CGF.CGM.getOpenMPRuntime().initLastprivateConditionalCounter(CGF, S);
CGF.EmitOMPLoopBody(S, LoopExit);
CGF.EmitStopPoint(&S);
}
@@ -2011,8 +2024,6 @@
CGF.EmitOMPInnerLoop(
S, LoopScope.requiresCleanups(), S.getCond(), S.getInc(),
[&S](CodeGenFunction &CGF) {
- CGF.CGM.getOpenMPRuntime().initLastprivateConditionalCounter(
- CGF, S);
CGF.EmitOMPLoopBody(S, CodeGenFunction::JumpDest());
CGF.EmitStopPoint(&S);
},
@@ -2667,8 +2678,6 @@
: S.getCond(),
StaticChunkedOne ? S.getDistInc() : S.getInc(),
[&S, LoopExit](CodeGenFunction &CGF) {
- CGF.CGM.getOpenMPRuntime()
- .initLastprivateConditionalCounter(CGF, S);
CGF.EmitOMPLoopBody(S, LoopExit);
CGF.EmitStopPoint(&S);
},
@@ -2851,7 +2860,6 @@
// break;
// }
// .omp.sections.exit:
- CGF.CGM.getOpenMPRuntime().initLastprivateConditionalCounter(CGF, S);
llvm::BasicBlock *ExitBB = CGF.createBasicBlock(".omp.sections.exit");
llvm::SwitchInst *SwitchStmt =
CGF.Builder.CreateSwitch(CGF.EmitLoadOfScalar(IV, S.getBeginLoc()),