tianshilei1992 created this revision.
Herald added subscribers: cfe-commits, guansong, yaxunl.
Herald added a reviewer: jdoerfert.
Herald added a project: clang.
Previously for nowait target, CG emitted a function call to
`__tgt_target_nowait`, etc. However, in OpenMP RTL, these functions just
directly call the no-nowait version, which means nowait is not working as
expected.
OpenMP specification says a target is acutally a target task, which is an untied
and detachable task. It is natural to go to the direction that generates a task
for a nowait target. However, OpenMP task has a problem that it must be within
to a parallel region; otherwise the task will be executed immediately. As a
result, if we directly wrap to a regular task, the nowait target outside of a
parallel region is still a synchronous version.
In D77609 <https://reviews.llvm.org/D77609>, I added the support for unshackled
task in OpenMP RTL. Basically,
unshackled task is a task that is not bound to any parallel region. So all
nowait target will be tranformed into an unshackled task. In order to
distinguish from regular task, a new flag bit is set for unshackled task. This
flag will be used by RTL for later process.
Signed-off-by: Shilei Tian <tianshilei1...@gmail.com>
Repository:
rG LLVM Github Monorepo
https://reviews.llvm.org/D78075
Files:
clang/lib/CodeGen/CGOpenMPRuntime.cpp
clang/lib/CodeGen/CGOpenMPRuntime.h
clang/lib/CodeGen/CGStmtOpenMP.cpp
Index: clang/lib/CodeGen/CGStmtOpenMP.cpp
===================================================================
--- clang/lib/CodeGen/CGStmtOpenMP.cpp
+++ clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -3615,6 +3615,8 @@
OMPTaskDataTy Data;
// The task is not final.
Data.Final.setInt(/*IntVal=*/false);
+ // The task is unshackled
+ Data.Unshackled = true;
// Get list of firstprivate variables.
for (const auto *C : S.getClausesOfKind<OMPFirstprivateClause>()) {
auto IRef = C->varlist_begin();
Index: clang/lib/CodeGen/CGOpenMPRuntime.h
===================================================================
--- clang/lib/CodeGen/CGOpenMPRuntime.h
+++ clang/lib/CodeGen/CGOpenMPRuntime.h
@@ -118,6 +118,7 @@
unsigned NumberOfParts = 0;
bool Tied = true;
bool Nogroup = false;
+ bool Unshackled = false;
};
/// Class intended to support codegen of all kind of the reduction clauses.
Index: clang/lib/CodeGen/CGOpenMPRuntime.cpp
===================================================================
--- clang/lib/CodeGen/CGOpenMPRuntime.cpp
+++ clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -5045,6 +5045,7 @@
DestructorsFlag = 0x8,
PriorityFlag = 0x20,
DetachableFlag = 0x40,
+ UnshackledFlag = 0x80,
};
unsigned Flags = Data.Tied ? TiedFlag : 0;
bool NeedsCleanup = false;
@@ -5057,6 +5058,8 @@
Flags = Flags | PriorityFlag;
if (D.hasClausesOfKind<OMPDetachClause>())
Flags = Flags | DetachableFlag;
+ if (Data.Unshackled)
+ Flags |= UnshackledFlag;
llvm::Value *TaskFlags =
Data.Final.getPointer()
? CGF.Builder.CreateSelect(Data.Final.getPointer(),
@@ -9789,6 +9792,7 @@
assert(OutlinedFn && "Invalid outlined function!");
const bool RequiresOuterTask = D.hasClausesOfKind<OMPDependClause>();
+ const bool HasNowaitClause = D.hasClausesOfKind<OMPNowaitClause>();
llvm::SmallVector<llvm::Value *, 16> CapturedVars;
const CapturedStmt &CS = *D.getCapturedStmt(OMPD_target);
auto &&ArgsCodegen = [&CS, &CapturedVars](CodeGenFunction &CGF,
@@ -9801,7 +9805,8 @@
llvm::Value *MapTypesArray = nullptr;
// Fill up the pointer arrays and transfer execution to the device.
auto &&ThenGen = [this, Device, OutlinedFn, OutlinedFnID, &D, &InputInfo,
- &MapTypesArray, &CS, RequiresOuterTask, &CapturedVars,
+ &MapTypesArray, &CS, RequiresOuterTask, HasNowaitClause,
+ &CapturedVars,
SizeEmitter](CodeGenFunction &CGF, PrePostActionTy &) {
if (Device.getInt() == OMPC_DEVICE_ancestor) {
// Reverse offloading is not supported, so just execute on the host.
@@ -9851,7 +9856,6 @@
// Emit tripcount for the target loop-based directive.
emitTargetNumIterationsCall(CGF, D, DeviceID, SizeEmitter);
- bool HasNowait = D.hasClausesOfKind<OMPNowaitClause>();
// The target region is an outlined function launched by the runtime
// via calls __tgt_target() or __tgt_target_teams().
//
@@ -9897,9 +9901,7 @@
NumTeams,
NumThreads};
Return = CGF.EmitRuntimeCall(
- createRuntimeFunction(HasNowait ? OMPRTL__tgt_target_teams_nowait
- : OMPRTL__tgt_target_teams),
- OffloadingArgs);
+ createRuntimeFunction(OMPRTL__tgt_target_teams), OffloadingArgs);
} else {
llvm::Value *OffloadingArgs[] = {DeviceID,
OutlinedFnID,
@@ -9908,10 +9910,8 @@
InputInfo.PointersArray.getPointer(),
InputInfo.SizesArray.getPointer(),
MapTypesArray};
- Return = CGF.EmitRuntimeCall(
- createRuntimeFunction(HasNowait ? OMPRTL__tgt_target_nowait
- : OMPRTL__tgt_target),
- OffloadingArgs);
+ Return = CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__tgt_target),
+ OffloadingArgs);
}
// Check the error code and execute the host version if required.
@@ -9923,7 +9923,7 @@
CGF.Builder.CreateCondBr(Failed, OffloadFailedBlock, OffloadContBlock);
CGF.EmitBlock(OffloadFailedBlock);
- if (RequiresOuterTask) {
+ if (RequiresOuterTask || HasNowaitClause) {
CapturedVars.clear();
CGF.GenerateOpenMPCapturedVars(CS, CapturedVars);
}
@@ -9934,10 +9934,9 @@
};
// Notify that the host version must be executed.
- auto &&ElseGen = [this, &D, OutlinedFn, &CS, &CapturedVars,
- RequiresOuterTask](CodeGenFunction &CGF,
- PrePostActionTy &) {
- if (RequiresOuterTask) {
+ auto &&ElseGen = [this, &D, OutlinedFn, &CS, &CapturedVars, RequiresOuterTask,
+ HasNowaitClause](CodeGenFunction &CGF, PrePostActionTy &) {
+ if (RequiresOuterTask || HasNowaitClause) {
CapturedVars.clear();
CGF.GenerateOpenMPCapturedVars(CS, CapturedVars);
}
@@ -9945,7 +9944,7 @@
};
auto &&TargetThenGen = [this, &ThenGen, &D, &InputInfo, &MapTypesArray,
- &CapturedVars, RequiresOuterTask,
+ &CapturedVars, RequiresOuterTask, HasNowaitClause,
&CS](CodeGenFunction &CGF, PrePostActionTy &) {
// Fill up the arrays with all the captured variables.
MappableExprsHandler::MapBaseValuesArrayTy BasePointers;
@@ -10035,15 +10034,16 @@
Address(Info.PointersArray, CGM.getPointerAlign());
InputInfo.SizesArray = Address(Info.SizesArray, CGM.getPointerAlign());
MapTypesArray = Info.MapTypesArray;
- if (RequiresOuterTask)
+ if (RequiresOuterTask || HasNowaitClause)
CGF.EmitOMPTargetTaskBasedDirective(D, ThenGen, InputInfo);
else
emitInlinedDirective(CGF, D.getDirectiveKind(), ThenGen);
};
- auto &&TargetElseGen = [this, &ElseGen, &D, RequiresOuterTask](
- CodeGenFunction &CGF, PrePostActionTy &) {
- if (RequiresOuterTask) {
+ auto &&TargetElseGen = [this, &ElseGen, &D, RequiresOuterTask,
+ HasNowaitClause](CodeGenFunction &CGF,
+ PrePostActionTy &) {
+ if (RequiresOuterTask || HasNowaitClause) {
CodeGenFunction::OMPTargetDataInfo InputInfo;
CGF.EmitOMPTargetTaskBasedDirective(D, ElseGen, InputInfo);
} else {
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
